Design and analysis of interactive multi-robot systems without explicit communication
- Zijian Wang.
- [Stanford, California] : [Stanford University], 2019.
- Copyright notice
- Physical description
- 1 online resource.
Also available at
- Wang, Zijian, author.
- Schwager, Mac, degree supervisor.
- Pavone, Marco, 1980- degree committee member.
- Rock, Stephen M., degree committee member.
- Stanford University. Department of Aeronautics and Astronautics.
- ["The mission of this thesis is to develop decentralized control algorithms that enable intelligent robots to interact with other robots or humans without explicit communication. Study of the interactive behaviors among robots will play an important role for our future society as we expect to have an increasingly large number of robots to serve in people's everyday lives. Unfortunately, most robots today lack awareness and understanding of other surrounding robots or humans and have to be caged or fenced, for example in factories, in order to operate safely. In this thesis, we develop a series of scalable approaches to address how a large swarm of robots cooperate and compete with each other. The first part of this thesis deals with multi-robot manipulation, a cooperative task where a group of robots collaborate to move a common heavy payload. We first propose a decentralized force and torque controller for planar object manipulation that allows many follower robots to synchronize their wrenches to a leader robot, who guides the entire team, using the measurement of the object's motion at their local attachment points. We rigorously prove the convergence guarantee of the controller using nonlinear control theory and Lyapunov style stability arguments. We verify the proposed method in simulations with up to 1000 robots to move various objects such as a simulated piano of realistic size. Two custom robot platforms, the differential-drive m3Pi robot and the omni-directional OuijaBot, are designed and built for experimental validation under real-world noises and disturbances. Next, we extend our planar manipulation to the 3-D space by rigidly attaching a group of quadrotor aerial robots to the payload. We propose a decentralized wrench coordination algorithm that can mathematically guarantee the 6-degree-of-freedom stability of the payload despite the challenge arising from motor saturation. We also provide a trajectory planning pipeline which explicitly accounts for the force/torque input constraint and therefore guarantees feasibility. Both simulations and experiments with quadrotors are performed to show the effectiveness of our approach for multi-robot manipulation in 3-D. The second part of this thesis addresses competition in the context of multi-robot racing. We present a real-time game theoretic planning algorithm for a robotic vehicle to race against multiple opponents on a racecourse. Using a novel iterative best response algorithm, we can find approximate Nash equilibria in the space of the multiple robots' trajectories that account for the opponents' intentions and responses. We also apply machine learning techniques to learn from the historical racing data offline, which can then be used to accelerate and improve the online solution during real-time planning. We conduct extensive simulation studies, and statistics reveal that our game theoretic planner largely outperforms a baseline model predictive controller that does not consider the opponents' responses. Experiments are conducted with four quadrotor aerial robots to validate our approach in real time and with physical robot hardware. All of our proposed algorithms have the feature of using no explicit communication among robots. In multi-robot manipulation, we show that local sensing of the payload's state of motion can acquire enough information for global coordination. In game theoretic planning, robots account for competitors' reactions through intention inference. This feature greatly extends the usability of our approaches when a communication network is unavailable or unreliable among a large swarm of robots."]
- Publication date
- Copyright date
- Submitted to the Department of Aeronautics and Astronautics.
- Thesis Ph.D. Stanford University 2019.