%{search_type} search results

The last decade saw the rapid development of two major mobility paradigms: Mobility-on-Demand (MoD) systems (e.g. ridesharing, carsharing) and self-driving vehicles. While individually impactful, together they present a major paradigm shift in modern mobility. Autonomous Mobility-on-Demand (AMoD) systems, wherein a fleet of self-driving vehicles serve on-demand travel requests, present a unique opportunity to alleviate many of our transportation woes. Specifically, by combining fully-compliant vehicles with central coordination, AMoD systems can achieve system-level optimal strategies via, e.g., coordinated routing and preemptive dispatch. This thesis presents methods to model, analyze and control AMoD systems. In particular, special emphasis is given to develop stochastic algorithms that can cope with the uncertainty inherent to travel demand. In the first part, we present a steady-state modeling framework built on queueing networks and network flow theory. By casting the system as a multi-class BCMP network, the framework provides analysis tools that allow the characterization of performance metrics for a given routing policy, in terms, e.g., of vehicle availabilities, and first and second order moments of vehicle throughput. Moreover, we present a scalable method for the synthesis of routing policies, with performance guarantees in the limit of large fleet sizes. The framework provides a large set of modeling options, and specifically address cases where the operational concerns of congestion and battery charge level are considered. We validate our theoretical results on a case study of New York City. In the second part, we leverage the insights provided by the steady-state models to present real-time control algorithms. Specifically, we cast the real-time control problem within a stochastic model predictive control framework. The control loop consists of a forecasting generative model and a stochastic optimization subproblem. At each time step, the generative model first forecasts a finite number of travel demand for a finite horizon and then we solve the stochastic subproblem via Sample Average Approximation. We show via simulation that this approach is more robust to uncertain demand and vastly outperforms state-of-the-art fleet-level control algorithms. Finally, we validate the presented frameworks by deploying a fleet control application in a carsharing system in Japan. The application uses the aforementioned algorithms to provide, in real-time, tasks to the carsharing employees regarding actions to be taken to better meet customer demand. Results show significant improvement over human based decision making.

Transportation is a necessary resource for many societies around the world. While advances in data science provide promising tools for personalized, adaptive and more efficient mobility services, they also bring new challenges in equal measure. In this dissertation I will discuss algorithm design for two such challenges faced by modern mobility services. First, I will discuss techniques for operating ridehailing and ridesharing systems in settings with incomplete information, which often arise due to the on-demand nature of such services. In particular, I will show both in theory and in practice how ideas from model predictive control, online optimization and machine learning can be used to effective serve existing customers while also adequately preparing for unknown future demand. Second, I will highlight some privacy concerns that arise from the sharing of mobility data that is often required for modern data-driven algorithms. To address some of these concerns, I present techniques based on multiparty computation and differential privacy to effectively use location data to improve routing services in a privacy-preserving way

Autonomous vehicles have the capability to revolutionize human mobility and vehicle safety. To prove safe, they must be capable of navigating their environment as well as or better than the best human drivers. The best human drivers can leverage the limits of a vehicle's capabilities to avoid collisions and stabilize the vehicle while sliding on pavement, ice, and snow. Automated vehicles should similarly be capable of navigating safety-critical scenarios when friction is limited, and one large advantage they hold over human drivers is the amount of data they can generate. With self-driving vehicles in the San Francisco Bay Area collecting almost two human lifetimes worth of data just during 2020, this abundance of data holds the key to improving vehicle safety. This dissertation examines how data generated by self-driving vehicles can be used to learn control policies and models to improve vehicle control near the limits of handling. As data collection and vehicle operation near the limits can be expensive, this work uses skilled humans as an inspiration for learning policies because of their incredible data efficiency. This ability is clearly demonstrated in racing where skilled human drivers act to improve their performance after each lap by shifting their braking point to maximize corner entry speed and minimize lap time. Starting from a benchmark feedforward and feedback control architecture already comparable to skilled human drivers, this work directly learns feedforward policies to improve vehicle performance over time. By using an approximate physics-based model of the vehicle, recorded lap data, and the gradient of lap time, this approach improves lap time by almost seven tenths of a second on a nineteen second lap over an initial optimization-based approach for racing. Additionally, this approach generalizes to low-friction driving. While model-based policy search shows improvement over a solely optimization-based approach, model-based policy search is ultimately limited by the vehicle model used. Physics-based models are useful for interpretability and understanding, but fail to make use of the abundance of data self-driving vehicles generate and often do not capture high-order or complex-to-model effects. Additionally, to operate at a vehicle's true limits, precise identification of the vehicle's road-tire friction coefficient is required which is a very difficult task. To overcome the drawbacks of physics-based models, this thesis next examines the ability of neural networks to use vehicle data to learn vehicle dynamics models. These models are capable of not only modeling higher-order and complex effects, but also vehicle motion on high- and low-friction surfaces. Furthermore, these models do so while retaining comparable control performance near the limits to a benchmark physics-based feedforward and feedback control architecture. Though this control approach shows promise in operating near the limits, feedforward and feedback control is ultimately limited in its ability to trade of small errors in the short term to prevent larger errors in the future. Additionally, actuator and road boundary constraints play an increasingly important role in safety as the vehicle nears the limits. To deal with these limitations, this work presents neural network model predictive control for automated driving near the limits of friction. Neural network model predictive control not only leverages the neural network model's ability to predict dynamics on high- and low-friction test tracks, but also retains comparable or better performance to MPC using a well-tuned physics model optimized to the corresponding high- or low-friction test track. While neural network MPC shows improved performance over physics-based MPC when operating near the limits, MPC leverages its dynamics model with complete certainty. These effects can lead to MPC overleveraging its dynamics model, which in the presence of model mismatch can lead to poor controller performance. Additionally, when using neural network models in MPC, the network predicts vehicle motion with complete certainty regardless of the presence or absence of training data in the corresponding modeled region. To mitigate this issue, this work presents an approach which leverages a neural network model to learn the uncertainty in the underlying dynamics model used in MPC. By learning the uncertainty in MPC's dynamics model, the vehicle can take actions to avoid highly uncertain regions of operation while still attempting to optimize the original MPC cost function. The insights from this work can be used to design automated vehicles capable of leveraging vehicle data to more effectively operate near the limits of handling

Autonomous systems are increasingly nearing widespread adoption, with new robotic platforms constantly being tested and deployed alongside humans in domains such as autonomous driving, service robotics, and surveillance. Accordingly, human-robot interaction will soon be present in many everyday scenarios. However, there are still many challenges preventing autonomous systems from safely and smoothly navigating interactions with humans. For example, while merging into traffic is one of the most common day-to-day maneuvers we perform as drivers, it poses a major problem for state-of-the-art self-driving vehicles. The reason humans can naturally navigate through many social interaction scenarios, such as merging in traffic, is that humans have an intrinsic capacity to reason about other people's intents, beliefs, and desires, applying this reasoning to predict what might happen in the future and make corresponding decisions. As a result, imbuing autonomous systems with the ability to reason about other agents' potential future actions is critical to enabling informed decision making and proactive actions to be taken in human-robot interaction scenarios. Indeed, the ability to predict other agents' behaviors (also known as "trajectory forecasting") has already become a core component of modern robotic systems, especially so in safety-critical applications such as autonomous vehicles. Towards this end, this dissertation tackles the development of trajectory forecasting methods, their effective integration within the robotic autonomy stack, and the injection of task-awareness in their performance evaluation

Augmenting robotic intelligence with cloud connectivity is considered one of the most promising solutions to cope with growing volumes of rich robotic sensory data and increasingly complex perception and decision-making tasks. While the benefits of cloud robotics have been envisioned long before, we have historically lacked flexible methods to trade-off the benefits of cloud computing with end-to-end systems costs of network delay, cloud storage, human annotation time, and cloud-computing time. To address this need, this thesis introduces decision-theoretic algorithms that allow robots to significantly transcend their on-board perception capabilities by using cloud computing, but in a low-cost, fault-tolerant manner. The utility of these algorithms is demonstrated on months of field data and experiments on state-of-the-art embedded deep learning hardware. Specifically, for compute-and-power-limited robots, this thesis presents a lightweight model selection algorithm that learns when a robot should exploit low-latency on-board computation, or, when highly uncertain, query a more accurate cloud model. Then, I present a collaborative learning algorithm that allows a diversity of robots to mine their real-time sensory streams for valuable training examples to send to the cloud for model improvement. This thesis concludes by surveying a number of future research directions on the systems and theoretical aspects of networked system control, some of which extend beyond cloud robotics

In order to keep self-driving cars safe, it is important that these vehicles can plan safe trajectories through their environment and have the ability to robustly use their full tire-force potential. Racing at the limits of handling provides a purposefully challenging scenario for the development of reliable vehicle-motion planning and control techniques, as race cars are constantly pushed to their physical limits. With a common trajectory-tracking architecture for automated vehicle control, steering provides path-tracking control, and the throttle and brakes are used to track a desired speed profile. For the specific application of racing, this speed profile can be designed to fully use the tire-force potential. Experimental data show that a preexisting control framework based on this approach can match the lap time of an amateur race-car driver, but a professional race-car driver proves to be slightly faster. It is demonstrated with both experimental results and an analytical method that with this decoupled path-tracking and speed-tracking controller, an automated vehicle is prone to either under-utilize the tires or lose control over the path-tracking dynamics when unintentionally operating beyond the limit. Furthermore, a professional race-car driver successfully operates the vehicle with a control strategy that seems fundamentally different from trajectory tracking. Namely, he shows significant lap-to-lap variations in both speed and path, but he is consistently faster than automated trajectory-tracking control. This inspires new strategies for automated vehicle control. In this context, two novel feedback-control strategies are presented, which employ slip-angle control to robustly use the vehicle's full tire-force potential, while speed control provides the path-tracking functionality. Subsequently, in order to have the ability to also adjust the vehicle's path, a Nonlinear Model Predictive Control (NMPC) framework is presented which can trade-off longitudinal and lateral control inputs. Experimental results demonstrate that this controller successfully coordinates the inputs at the limits of handling. However, the computational burden of this NMPC framework limits the length of the planning horizon for real-time control, which in turn inhibits its ability to adjust the vehicle's path and speed. To address this issue, a new NMPC framework is developed, which serially cascades vehicle models of different levels of complexity in the planning horizon. Experimental results on an automated race car demonstrate the benefits of this new concept, with a high quality of control provided by a high-fidelity vehicle model in the near-term planning horizon, and significant extension of the planning horizon with a low-fidelity model.

Integrating autonomous robots into safety-critical settings requires reasoning about uncertainty at all levels of the autonomy stack. This thesis presents novel algorithmic tools for imbuing robustness within two hierarchically complementary areas, namely: motion planning and decision-making. In Part I of the thesis, by harnessing the theories of contraction and semi-infinite convex optimization and the computational tool of sum-of-squares programming, we present a unified framework for robust real-time motion planning for complex underactuated nonlinear systems. Broadly, the approach entails pairing open-loop motion planning algorithms that neglect uncertainty and are optimized for generating trajectories for simple kinodynamic models in real-time, with robust nonlinear trajectory-tracking feedback controllers. We demonstrate how to systematically synthesize these controllers and integrate them within planning to generate and execute certifiably safe trajectories that are robust to the closed-loop effects of disturbances and planning with simplified models. In Part II of the thesis, we demonstrate how to embed the control-theoretic advancements developed in Part I as constraints within a novel semi-supervised algorithm for learning dynamical systems from user demonstrations. The constraints act as a form of context-driven hypothesis pruning to yield learned models that jointly balance regression performance and stabilizability, ultimately resulting in generated trajectories for the robot that are conditioned for feedback control. Experimental results on a quadrotor testbed illustrate the efficacy of the proposed algorithms in Parts I and II of the thesis, and clear connections between theory and hardware. Finally, in Part III of the thesis, we describe a framework for lifting notions of robustness from low-level motion planning to higher-level sequential decision-making using the theory of risk measures. Leveraging a class of risk measures with favorable axiomatic foundations, we demonstrate how to formulate decision-making algorithms with tunable robustness properties. In particular, we focus on a novel application of this framework to inverse reinforcement learning where we learn predictive motion models for humans in safety-critical scenarios, and illustrate their effectiveness within a commercial driving simulator featuring humans in-the-loop. The contributions within this thesis constitute an important step towards endowing modern robotic systems with the ability to systematically and hierarchically reason about safety and efficiency in the face of uncertainty, which is crucial for safety-critical applications.

This thesis is about the design and analysis of robots that use adhesives to combine multiple modes of operation. Examples of multimodal operation include ballistic or powered flight combined with perching and crawling. In each case, the robots are made possible by a proliferation of components -- from microprocessors to sensors and motors -- that have accompanied the growth of drones or quadrotors in consumer markets. The components are compact and light enough that it is possible to support multiple modes of operation on a small platform. The thesis takes a cue from small creatures such as insects, most of which have multiple modes of operation (e.g. flying and crawling) and which can often move objects many times their weight through the use of attachment mechanisms at the tarsus of each limb. Examining the strategies of insects such as wasps leads to insights for designing and modeling the multimodal robots outfitted with gecko-inspired adhesives. The resulting platforms are capable of tasks that no single mode of operation can support. For example, one of these platforms, named FlyCroTugs, can fly rapidly to remote sites, attach a tether to a heavy object, land and then pull that object with a force many times the robot's weight. The FlyCroTugs use gecko-inspired adhesives to anchor themselves when applying large forces. Another example involves a small robot, named KlingOn, that can transition from ballistic flight to crawling on a vertical surface, using the same gecko-inspired adhesives. A third example involves a gripper that can capture free-floating objects with a flexible-backed adhesives. Further insights arise when considering scaling laws applicable to small multimodal robots that use adhesion. At the scale of these robots, contact forces typically dominate the dynamics when the robots are interacting with objects. Therefore, modeling the force constraints associated with the adhesives leads to corresponding dynamic constraints on the robots, in terms of their trajectories and velocities. The adhesive force constraints also have implications for the dimensions, geometry, stiffness and damping of the robot attachment pads and grippers. Ultimately this thesis posits that increased attention to robotic end-effectors and attachment mechanisms can promote the efficacy of small, multimodal robotic systems interacting with their environment.

Spins are upset maneuvers in which an asymmetric stall over an airplane's wing causes it to enter a steep downward helical trajectory, often with reduced, annulled, or reversed control surface effectiveness. If these occur at low altitude, there might not be enough airspace to recover before colliding with the ground. Historically, this hazard has been addressed by careful aerodynamic design to suppress or minimize spin tendencies, and by flight crew training. Despite major reductions in accident rates, improvements have stagnated in recent decades, requiring new approaches to the problem. This dissertation proposes a software enabled approach, developing algorithms that can detect spins at an early stage and automatically recover with minimal altitude loss. To enable this study, a high angle of attack aerodynamic model of a typical general aviation aircraft is identified from wind tunnel and flight data. Using this model, the minimal altitude optimal control problem is investigated, and a spin recovery controller is designed. In addition, the relation between arrest delay and altitude loss is quantified, showing that altitude loss grows rapidly within the first turn. Motivated by these results, a methodology for designing spin detection schemes using different sensors is proposed. The methodology is applied to the same general aviation aircraft showing that detection at an early stage of the incipient phase is possible, resulting in as much as a fourfold reduction in altitude loss with respect to recovery from one-turn spins by a human pilot. Finally, the spin detection and recovery system is tested on small-scale UAVs, demonstrating the predicted fourfold altitude loss reduction. The results obtained indicate that such a system could help reduce spin-related accident rates by as much as 45%.

Grasping is central to robotics, and numerous hands and grippers have been developed for applications ranging from personal service robots to industrial manufacturing lines. Either by wrapping around an object to support it from underneath or by using internal forces and friction, these traditional grippers can grasp many objects securely. However, this approach does not work when the object is much larger than the hand. In this case, adhesion becomes a solution. Various adhesive or astrictive technologies have been applied in robotics, including suction, magnetic forces and electrostatic forces. Most of these technologies require active control and consume power. Many are also limited to a range of surface properties or materials. In nature, we find alternative solutions including arrays of small spines for attaching to rough surfaces and structures with microscopic features that can attach to smooth surfaces using van der Waals forces. For example, many insects use tiny spines and claws on their legs to catch on small bumps and pits (asperities) on rough surfaces for climbing and grasping. Gecko lizards have demonstrated van der Waals adhesion on smooth surfaces using fine hairs, or setae, on the feet. These types of adhesion do not require power and have been proven strong, reusable, controllable, and adaptive to a wide range of surfaces. Learning from biology and using bio-inspired adhesion can facilitate robust grasping of large objects. Previous work has considered bio-inspired adhesion for climbing robots, where shear loads dominate. This thesis focuses on adhesion for grasping and manipulation so that the robot can apply forces and moments in any direction. The work expands the modeling and analysis of bio-inspired, directional adhesion, introduces an opposed-grip strategy for grasping and manipulation, and presents efficient scaling methods to accommodate large forces and moments with special considerations for dynamic impacts. For insect-inspired microspines, a statistical model describes the asperity spatial and strength distributions and predicts the shear and normal adhesion capabilities of a group of microspines. Various opposed-grip mechanisms are designed and modeled for grasping a range of rough surfaces such as concrete, asphalt and the bark of trees. The stochastic models of spine/surface interaction lead to insights for designing grippers that work on a range of surfaces. A new opposed-spine gripper is demonstrated on a small aerial platform, allowing it to perch on building walls and ceilings to save power and greatly extend mission life. For gecko-inspired adhesives, similar opposed mechanisms are developed for both flat and curved surfaces. Based on the insights obtained from modeling arrays of directional adhesive tiles or films, a series of grippers are shown to have significantly increased adhesion capabilities as compared to simply increasing the area for a single adhesive patch. For large loads, different scaling strategies, such as a pulley differential and constant force springs, are developed and compared for various applications. Outriggers are added to grippers to enlarge moment capabilities for manipulation. A passive, nonlinear wrist, in conjunction with scaled grippers, is presented to facilitate efficient energy absorption during dynamic impacts while also remaining stiff for precise manipulation and providing overload protection for regular operations. An integrated gripper with the above elements demonstrates grasping and manipulation of objects several times larger than the gripper itself. The limiting force and moment space of such a gripper can be computed, providing useful information for robotic motion planning and control. A series of experiments on floating platforms, conducted with the NASA Jet Propulsion Laboratory, demonstrates the applicability of this approach for grasping and manipulating objects in space.

This thesis presents a full-stack, real-time planning framework for kinodynamic robots that is enabled by a novel application of machine learning for reachability analysis. As products of this work, three contributions are discussed in detail in this thesis. The first contribution is the novel application of machine learning for rapid approximation of reachable sets for dynamical systems. The second contribution is the synthesis of machine learning, sampling-based motion planning, and optimal control into a cohesive planning framework that is built on an offline-online computation paradigm. The final contribution is the application of this planning framework on a quadrotor system to produce, arguably, one of the first demonstrations of fully-online kinodynamic motion planning. During physical experiments, the framework is shown to execute planning cycles at a rate 3 Hz to 5 Hz, a significant improvement over existing techniques. For the quadrotor, a simplified dynamics model is used during the planning phase to accelerate online computation. A trajectory smoothing phase, which leverages the differentially flat nature of quadrotor dynamics, is then implemented to guarantee a dynamically feasible trajectory. An event-based replanning structure is implemented to handle the case of dynamic, even adversarial, obstacles. A locally reactive control layer, inspired by potential fields methods, is added to the framework to help minimizes replanning events and produce graceful avoidance maneuvers in the presence of high speed obstacles.

As autonomous vehicles enter public roads, they should be capable of using all of the vehicle's performance capability, if necessary, to avoid collisions. This dissertation focuses on facilitating collision avoidance for autonomous vehicles by enabling safe vehicle operation up to the handling limits. The new control approaches first rely on a standard paradigm for autonomous vehicles that divides vehicle control into trajectory generation and trajectory tracking. A trajectory generation approach calculates emergency lane change trajectories, defined in terms of path curvature, that allows an autonomous vehicle to perform emergency lane changes up to its handling limits. Analysis also provides insights into when and to what extent a vehicle should brake and turn during an emergency lane change to maximize the number of situations in which a collision can be avoided. However, experimental results also highlight vehicle stabilization challenges associated with tracking paths defined by high rates of curvature change, which are desirable for emergency maneuvers. A link is forged between path curvature and vehicle performance, which inspires two trajectory tracking control designs. A four-wheel steering controller adds rear steering actuation to improve tracking and stabilization performance, while a two-wheel steering predictive controller incorporates future path information into current control actions. Experimental results demonstrate the advantages of each approach. However, separating vehicle control into trajectory generation and tracking is not always conducive to emergency maneuvers up to the vehicle's handling limits, where these aspects of vehicle control become tightly coupled with each other and with vehicle stabilization. An alternative paradigm is suggested that is more adept at controlling the vehicle in such scenarios. This approach integrates trajectory generation, trajectory tracking, and vehicle stabilization into one controller capable of mediating among the sometimes conflicting demands imposed by collision avoidance and stabilization. The controller can prioritize collision avoidance, above even stabilization, to minimize potential collisions. Experimental emergency lane changes and a mid-corner obstacle avoidance scenario highlight the advantages of this integrated approach to vehicle control.

Professional drivers can control their vehicle's position, speed and orientation with an incredible amount of precision even when the open loop dynamics are unstable. They do so by leveraging their understanding of the vehicle's behavior and simultaneous coordination of the vehicle's inputs. As autonomous vehicles are developed, they should be able to drive just as well as professional drivers to ensure successful navigation of emergency scenarios involving unintended rear tire saturation on the road. With that goal, this thesis presents improvements to the existing state of the art of autonomous vehicle drift control supported by experiments using a heavily modified 1981 DMC DeLorean, MARTY. We first present a novel technique to achieve a fully actuated system while drifting. Drawing inspiration from professional drivers, we use brakes along with steering and drive torque simultaneously and convert the system into a fully actuated one. Equilibrium analysis leveraging this simultaneous actuation confirms the expansion of the existing drift equilibria for a given curvature from a curve on the speed-sideslip plane to an area. Similarly, a tangent space analysis confirms that front braking adds another dimension to the available state derivatives; from a surface in the under-actuated case, to a volume. This represents a significant increase in the set of trajectories available to a drifting vehicle, and also an avenue to reliably reduce energy from the system while maintaining control of the vehicle's sideslip and position. Leveraging this fully actuated system, we then present an architecture capable of following a path, while tracking a desired sideslip and speed. The formulation builds upon the existing state of the are of drift control by incorporating speed tracking and precisely coordinating the three inputs. Experimental results using MARTY verify independent control over the position, speed and sideslip through a variety of trajectories while operating within the limits of actuation. They also highlight some limitations when operating at the limits of actuation. These limitations lead to the development of a novel application of nonlinear model predictive control for drifting. The model predictive control framework directly addresses the challenges of arbitrary error dynamics and actuator saturation as the vehicle model and actuation limits are easily embedded in the framework itself. The prediction horizon is capable of trading off between the different objectives when required similar to a professional driver. Experimental results illustrate significantly improved path tracking performance over the existing state of the art. They further demonstrate successful selective prioritization of different objectives. The controller can navigate difficult dynamic trajectories at the limits of actuation where the previous controller exhibited degraded performance. Finally, an augmented nonlinear model predictive control framework expands the operational domain of the vehicle to include the ability to track drift transitions. These highly dynamic maneuvers switch the direction of travel from left to right or vice versa and require operation in highly transient regions of the state space. Successful tracking of `Figure 8' trajectories verifies the expanded operational domain. Comparisons with the original formulation highlight the importance of the augmentations. Designing Figure 8 trajectories which leverage additional braking actuation has the potential to improve performance, however, requires additional model fidelity to be implemented. These contributions significantly increase the operational domain of a drifting vehicle, improve tracking performance, and mitigate the adverse effects of actuator saturation. They extend the applicability of drift controllers to common vehicles on the road which have much more limited actuation than purpose built drift cars. We hope that these contributions will form the basis of advanced safety systems and help improve vehicle safety

This dissertation develops a novel Dynamics, Guidance, and Control (DG&C) framework for multi-agent systems with specific application to so-called Distributed Space Systems (DSS). Spaceflight has being revolutionized by the usage of miniaturized low-size-weight-and-power (low-SWAP) satellites and commercial-off-the-shelf (COTS) technology. The trend makes access to space easier and more convenient for new and diverse stakeholders. At the same time, it enables the distribution of payload and tasks among multiple coordinated agents (i.e., DSS) enabling functionalities that are otherwise not achievable by single monolithic systems. The resulting novel DSS require autonomous and safe DG&C capabilities in a wide range of operational scenarios unmatched by previous spaceflight applications. After a review and harmonization of the mathematical foundation relevant to the DG&C of DSS, this research explores the method of variation of parameters (VoP) to conceive and develop a novel framework for optimal passively-safe control of multi-agent dynamics systems. Specifically, VoP is exploited to model the effects of non-integrable dynamics on the integration constants of an integrable portion of the governing dynamics itself. This fosters computational efficient dynamics modeling as well as inclusion of dynamics-dependent constraints with application to fault-tolerant control. Such advantages are used to address two specific challenges of the DG&C of DSS: 1) the accurate and efficient modeling of the complex relative motion dynamics in space; 2) the need of fuel and computationally efficient control algorithms capable of enforcing motion safety guarantees even in case of sudden loss of control capabilities by any agent, i.e., guarantees of passive safety. In fact, low-SWAP and COTS components reduce mission financial costs at the expense of reliability with higher risks of loss of control capabilities. This makes fault-tolerant motion safety particularly relevant, given the consequences a collision in space has in terms of debris generation and investments loss. The novel theoretical framework provides three main contributions to the state-of-the-art: 1) efficient inclusion of passive safety guarantees within a multi-agent optimal control problem solvable using direct methods, in presence of nonlinear non-integrable dynamics and realistic system uncertainties (from sensing, actuation, and unmodeled system dynamics); 2) novel closed-form linear dynamics models of the perturbed relative motion of DSS, for efficient on-board propagation, and adoption within DG&C algorithms; 3) novel closed-form solutions of passive safety for DSS, with contributions to relative orbit design in eccentric orbits and use within constrained optimal control problems. The main experimental contribution of this work is the application of the framework to the upcoming VIrtual Super Optics Reconfigurable Swarm (VISORS) mission, a first-of-a-kind nanosatellite segmented telescope due launch in 2024 with a 40 meters focal length for high-resolution imaging of the solar corona. Together with complementary challenging formation-flying test cases in eccentric orbits, the dissertation shows the advantages of the proposed approach in terms of achieved safety guarantees, control accuracy and gained fuel and computational efficiency

Advances in sensing and actuation capabilities have allowed for the proliferation of robots across many fields, including aerial, industrial, and automotive applications. A driving factor in being able to deploy such robots in everyday applications is algorithms that imbue real-time decision making capabilities. Such decision-making capabilities can be formulated using the modeling framework of optimization programs. However, such optimization-based approaches are still limited by computational resources available on robot platforms. For example, in many aerospace applications, spacecraft robotic systems are equipped with embedded computers much less capable than the hardware typically used to solve such optimization algorithms. Thus, there is a pressing need to be able to scale and extend optimization-based planning and control algorithms to robotics applications with severely constrained computational resources. In this work, we turn towards recent advances in nonlinear optimization, supervised learning, and control theory to accelerate solving optimization-based controllers for online deployment. We then show how data-driven approaches can exploit powerful computational resources offline to learn the underlying structure of optimization problems such that the online decision making problem can be reduced to an approximate problem that is much easier to solve on embedded computers. In the first part of this dissertation, we present a local trajectory optimization framework known as Guaranteed Sequential Trajectory Optimization (GuSTO) that provides a theoretically-motivated algorithm that iteratively solves a series of convex optimization problems until convergence. We demonstrate how this framework can accommodate a broad class of trajectory optimization problems, including free-final time, free final-state, and problems on a manifold. We further discuss how GuSTO enables new applications, specifically in the domain of spacecraft robotic manipulation, and discuss the development of a novel gecko-inspired adhesive robot gripper design for the Astrobee assistive free-flying robot. In the second part of this dissertation, we turn towards global trajectory optimization problems, specifically those that can be formulated as mixed-integer convex programs (MICPs). MICPs are a popular modeling framework that can be used to model planning and control problems that are inherently combinatorial or discrete. However, existing algorithms fall short in being able to provide reliable solution approaches that can be deployed for real-time applications (i.e., 10-100Hz computation rates) on embedded systems. In this work, we turn towards data-driven approaches that can be used to find high quality feasible solutions to such MICPs and present Combinatorial Offline, Convex Online (CoCo). We demonstrate how such approaches can leverage the underlying structure of optimal control problems and compare our proposed approach against state-of-the-art commercial solvers. Numerical simulations are provided through this work to demonstrate the efficacy of our proposed approach and present hardware results on a free-flying spacecraft robotic test bed

Advances in the fields of artificial intelligence and machine learning have unlocked a new generation of robotic systems---"learning-enabled" robots that are designed to operate in unstructured, uncertain, and unforgiving environments, especially settings where robots are required to interact in close proximity with humans. However, as learning-enabled methods, especially "deep" learning, continue to become more pervasive throughout the autonomy stack, it also becomes increasingly difficult to ascertain the performance and safety of these robotic systems and explain their behavior, necessary prerequisites for their deployment in safety-critical settings. This dissertation develops methods drawing upon techniques from the field of formal methods, namely Hamilton-Jacobi (HJ) reachability and Signal Temporal Logic (STL), to complement a learning-enabled robot autonomy stack, thereby leading to safer and more robust robot behavior. The first part of this dissertation investigates the problem of providing safety assurance for human-robot interactions, safety-critical settings wherein robots must reason about the uncertainty in human behavior to achieve seamless interactions with humans. Specifically, we develop a two-step approach where we first develop a learning-based human behavior prediction model tailored towards proactive robot planning and decision-making, which we then couple with a reachability-based safety controller that minimally intervenes whenever the robot is near safety violation. The approach is validated through human-in-the-loop simulation as well as on an experimental vehicle platform, demonstrating clear connections between theory and practice. The second part of this dissertation examines the use of STL as a formal language to incorporate logical reasoning into robot learning. In particular, we develop a technique, named STLCG, that casts STL into the same computational language as deep neural networks. Consequently, by using STLCG to express designers' domain expertise into a form compatible with neural networks, we can embed domain knowledge into learned components within the autonomy stack to provide additional levels of robustness and interpretability

Autonomous robots have the potential to free humans from dangerous or dull work. To achieve truly autonomous operation, robots must be able to understand unstructured environments and make safe decisions in the face of uncertainty and non-stationarity. As such, robots must be able to learn about, and react to, changing operating conditions or environments continuously, efficiently, and safely. While the last decade has seen rapid advances in the capabilities of machine learning systems driven by deep learning, these systems are limited in their ability to adapt online, learn with small amounts of data, and characterize uncertainty. The desiderata of learning robots therefore directly conflict with the weaknesses of modern deep learning systems. This thesis aims to remedy this conflict and develop robot learning systems that are capable of learning safely and efficiently. In the first part of the thesis we develop tools for efficient learning in changing environments. In particular, we develop tools for the meta-learning problem setting---in which data from a collection of environments may be used to accelerate learning in a new environment---in both the regression and classification setting. These algorithms are based on exact Bayesian inference on meta-learned features. This approach enables characterization of uncertainty in the face of small amounts of within-environment data, and efficient learning via exact conditioning. We extend these approaches to time-varying settings beyond episodic variation, including continuous gradual environmental variation and sharp, changepoint-like variation. In the second part of the thesis we adapt these tools to the problem of robot modeling and control. In particular, we investigate the problem of combining our neural network-based meta-learning models with prior knowledge in the form of a nominal dynamics model, and discuss design decisions to yield better performance and parameter identification. We then develop a strategy for safe learning control. This strategy combines methods from modern constrained control---in particular, robust model predictive control---with ideas from classical adaptive control to yield a computationally efficient, simple to implement, and guaranteed safe control strategy capable of learning online. We conclude the thesis with a discussion of short, intermediate, and long-term next steps in extending the ideas developed herein toward the goal of true robot autonomy

The last 20 years have seen considerable interest in bioinspired dry adhesives, based on discoveries regarding the adhesive system of the gecko and some arthropods. Such adhesives typically have the advantage of being reusable, leaving no residue, and allowing control of the adhesion through loading states. However, the number of practical applications of these adhesives remains small. One possible reason is that unlike in mechanical design, where design, simulation, and testing methodologies are all well established, there are significant gaps in all of these phases of engineering as applied to gecko-inspired adhesives. There are a variety of methods and metrics used for evaluating adhesives, often giving differing results, and even in some cases results that do not accurately reflect those observed in practical applications. Even with an accurate evaluation of an adhesive material, refining the design is challenging, as the design and manufacturing methods are typically time-consuming, highly constraining, or both. At the same time, there continues to be growing interest in the use of these adhesives in wide-ranging applications including reusable tapes and bandages; improved and more gentle industrial grippers; and grasping objects in space, where the combination of large objects, low contact forces, and lack of atmosphere make adhesives of particular interest. To address this growing need for improved ability to design and manufacture adhesives tailored to these applications, a three-pronged approach is taken. An improved method for testing gecko-inspired adhesives is presented. Unlike the common testing paradigms published in the literature, which impose a fixed displacement between the adhesive material and a test surface, the proposed testing method uses a series elastic configuration to apply forces to the adhesive. This shift in test control from displacement-space to force-space allows the testing conditions to be aligned to those seen in applications; whether for climbing, grasping, or adhesive tapes, nearly all applications of gecko-inspired adhesives fundamentally involve force-space constraints in normal conditions. It is shown that by testing the adhesives in similar conditions to those observed in use, the measured limit curves better reflect those seen in practice. Further, in cases where the adhesive structures are more complicated, or more integral to the performance of the adhesive---such as the directional, controllable adhesives at the core of this work---force-space testing enables measuring the full capabilities of the adhesive, which in many cases are impossible to measure in displacement-space. With the ability to accurately measure more complex limit curves, spatial variation is investigated as a means to improve the ability to create adhesives with novel parameters. In this case, the property of interest is a high friction ratio, the ratio of friction in a preferred direction to friction in the opposite direction, a property of the natural gecko adhesive system. Taking inspiration from the spatial variation found on the gecko's feet, an adhesive structure with wedges of varying length is developed, modeled, and analyzed. The friction ratio of this adhesive is measured, indicating an improvement of orders of magnitude over the current state of the art. Further, this adhesive structure also demonstrates the possibility of simplifying the adhesive design problem. Rather than developing a single complex feature to provide all of the desired properties, spatial variation permits the development of multiple features that are individually simpler but interact to provide more complex behavior. A discussion of the manufacturing process and associated fabrication constraints for these designed adhesive geometries follows. The process is an extension of a previous manufacturing process developed for making uniform adhesives. This is coupled with methods for directly incorporating adhesives into larger assemblies to create tightly coupled adhesive and sensing systems. Finally, a simplified design framework is presented, synthesizing many of the concepts from the prior sections. The current state of the art in adhesive simulation and modeling, while useful for understanding and explaining various specific aspects of adhesive design, is not adequate for directly analyzing the adhesion of complex adhesive geometries. The framework is intended to be a heuristic that synthesizes concepts from the various models of adhesion to provide useful guidance for thinking about adhesive designs for particular applications

Autonomous vehicles have the potential to greatly improve transportation safety by eliminating many automobile accidents, the vast majority of which are caused by human error. However, for cars to be able to avoid an accident whenever physically possible, they will have to drive at least as well as the best human drivers. Racing drivers can claim to be the best drivers in the world since, by the nature of their sport, they are forced to consistently and safely operate the vehicle at its physical limits. Autonomous racing provides an avenue to rapidly develop insights and control strategies for autonomous vehicles that are applicable to emergencies on public roads. This thesis expands the understanding of what effects must be captured for a vehicle to drive at the limits of friction. First, the impact of road topography on the vehicle's limits is discussed and modeled. Experiments with an automated vehicle show that accounting for topography-driven variation in normal load is critical for ensuring that the vehicle stays within its limits. The same simple model used to generate those insights is also useful for rapid trajectory replanning, illustrated here through examples covering obstacle avoidance and racing line optimization. This ap- proach to trajectory modification constitutes the second contribution of this thesis. While the simple model upon which the method is based captures the most funda- mental limitations of the vehicle, it is worth examining the extent to which more complex models of the vehicle's dynamics lead to better performance. An evaluation of the utility of several possible models for generating trajectories at the limit on various surfaces, including ice, wet asphalt, and dry asphalt, shows that the models' prescriptions for the optimal trajectory vary little and that all can be used success- fully. However, a significant advantage of the more complex models is that the many actuators available on modern vehicles may be used in a coordinated fashion to better accomplish the desired control objective. To this end, a novel model of the effects of a limited slip differential is incorporated into the double-track model of the vehi- cle. The insights from this work can be used to design algorithms that operate over the full range of vehicle performance, maximizing an autonomous vehicle's ability to operate skillfully when racing or safely when confronted with an emergency

From getting to work to strolling through the park, our mobility is an essential part of life. Losing one's mobility can be devastating. Scientists are on the verge of enhancing mobility for many movement disorders via exoskeletons. However, designing effective exoskeletons is challenging because of their tight coupling with the complex human body. Computer simulations of exoskeletons can reduce the duration of lengthy human experiments and reveal the effect of an exoskeleton on muscle coordination. A promising application for exoskeletons is reducing the burden of carrying heavy loads on the torso, which is a requirement of many occupations. To guide the design of such exoskeletons, my lab performed an experiment with seven male subjects walking while carrying 88 pounds on their torso. I used these data to simulate the effect of seven hypothetical idealized devices, each providing unrestricted torque at one joint in one direction (hip abduction, hip flexion, hip extension, knee flexion, knee extension, ankle plantarflexion, or ankle dorsiflexion). My simulations predicted that a device assisting with hip abduction would be most efficient at reducing the energy required to walk while carrying heavy loads. I found that many of our devices affected muscles that were not directly assisted. This result supported the notion that exoskeletons can have complex effects that are difficult to discover via experiments, or via simulations that do not include muscles. Although my simulations yielded valuable insights, I discovered that the method I employed limited the accuracy of my predictions. The method, named Computed Muscle Control, can optimize device torques and predict changes in muscle coordination but cannot predict changes to the walking motion itself. Musculoskeletal simulation tools usually model the nervous system via objectives we believe the brain minimizes. Even though individuals might employ different objectives for different motions, the nervous system objective that Computed Muscle Control employs cannot be modified. Lastly, Computed Muscle Control cannot optimize the values of constant model parameters, such as the stiffness of an assistive device. To address the limitations of Computed Muscle Control and related simulation tools, I created a flexible framework for optimizing the motion and control of musculoskeletal models. This framework, named Moco, employs the direct collocation method, which has become a popular approach for solving related problems within and beyond the field of biomechanics. Compared to other simulation tools, Moco provides an unprecedented amount of flexibility. Researchers can choose a nervous system objective from an existing library of modules. Moco is the first musculoskeletal direct collocation tool to handle kinematic constraints, which are common in musculoskeletal models. In collaboration with a labmate, I used Moco to design a passive device to assist with a squat-to-stand motion. We predicted the stiffness of the device and a new squat-to-stand motion without relying on motion data; such predictions were challenging to conduct with previous simulation tools. Moco will accelerate the use of simulations to predict the effect of exoskeletons, orthopedic surgeries, artificial joints, and other interventions that restore and enhance mobility