Skip navigation


1. Human-Robot Teaming.

A quadcopter is guided to pass through a window by a series of human's inputs. These inputs are given only when the performance of the robot does not meet human's expectation, i.e. in this case the quadrotor fails to pass through the window. The input is given in terms of directional corrections, for which magnitude of human's inputs does not matter any more. [Paper in IEEE TRO]. The demo is completed by Ph.D. student Zehui Lu.

2. Safe Navigation of Legged Robot

The demo is done by Ph.D. student Tianyu Zhou.


3. Multi-Agent Mission Planning

A distributed algorithm is developed to achieve multi-agent mission planning, i.e. the integration of task allocation (how to autonomously assign tasks to agents without conflict of interests) and path planning (how to autonomous guide all agents to their task locations). In the following video, two quadrotors communicate to decide the tasks (locations to visit) cooperatively, and then navigate to their task locations with collision-free path planning algorithms. The demo is done by Ph.D. students Zehui Lu and Tianyu Zhou.


4. Multi-Agent Formation Control and Learning

The demo is done by Ph.D. student Tianyu Zhou. The video shows a formation control of 3 quadrotors with Jackal UGV in the center. The quadrotors' dynamics are unknown. In the beginning, a random exploration noise is applied to both agents. The agents collect data and compute optimal control gain using computational adaptive dynamic programming. After the learning process, each agent updates the control inputs based on the real-time state information captured by the Motion Capture System. Based on the given formation information (relative states between following agents and leading agent) and the centroid of formation (Jackal's position), the agents make desired formation around Jackal.