CIS Department Talk - April 1, 2008
The Department of Computer and Information Science & The Society of
Computer Science Present
|Topic:||Real-Time Kinodynamic Planning: Physically-Realistic, Fast, Safe and Distributed
|Date:||April 1, 2008, 11:30AM|
|Place:||John Mulcahy Hall, Room 112|
Many exciting applications, ranging from realistic simulations and games to the control of robots, require real-time computation of collision-free trajectories given only partial knowledge of a potentially dynamic environment.
One promising and general approach to address such problems is to employ sampling-based kinodynamic planning, a paradigm that accommodates a variety of systems and directly addresses both geometric and dynamic aspects of motion planning. However, as a search-based approach, it poses computational challenges when time limitations are imposed for real-time performance. This results in low quality paths or partial paths that do not reach the goal. Moreover there are safety considerations, in terms of collision-avoidance, when a system has to respect dynamic motion constraints and operate under time limitations.
This talk describes three contributions in the context of real-time sampling-based kinodynamic planning. Firstly, we incorporate physical simulators in planning so as to be able to better represent realistic dynamics of the physical world, such as drift, friction, contacts and gravity. In this context, we work on "informed" versions of state-of-the-art planners, where any available workspace or domain information is utilized to reduce solution time, improve path quality and provide a high level guidance in the case of replanning. Secondly, we propose a "continuous" replanning approach that guarantees the safety of a system with dynamics in static environments with a reduced computational overhead. Finally, we have extended this solution to a distributed algorithm for multiple networked vehicles operating in the same environment. We show that through communication multiple vehicles can also achieve collision avoidance in real-time by employing sampling-based kinodynamic planners and the proposed safety framework.
For more information, contact Ms. Danielle Aprea (718) 817-4480