SAFETY-GUARANTEED TASK PLANNING FOR BIPEDAL NAVIGATION IN PARTIALLY OBSERVABLE ENVIRONMENTS
MetadataShow full item record
Bipedal robots are becoming more capable as basic hardware and control challenges are being overcome, however reasoning about safety at the task and motion planning levels has been largely underexplored. I would like to make key steps towards guaranteeing safe locomotion in cluttered environments in the presence of humans or other dynamic obstacles by designing a hierarchical task planning framework that incorporates safety guarantees at each level. This layered planning framework is composed of a coarse high-level symbolic navigation planner and a lower-level local action planner. A belief abstraction at the global navigation planning level enables belief estimation of non-visible dynamic obstacle states and guarantees navigation safety with collision avoidance. Both planning layers employ linear temporal logic for a reactive game synthesis between the robot and its environment while incorporating lower level safe locomotion keyframe policies into formal task specification design. The high-level symbolic navigation planner has been extended to leverage the capabilities of a heterogeneous multi-agent team to resolve environment assumption violations that appear at runtime. Modifications in the navigation planner in conjunction with a coordination layer allow each agent to guarantee immediate safety and eventual task completion in the presence of an assumption violation if another agent exists that can resolve said violation, e.g. a door is closed that another dexterous agent can open. The planning framework leverages the expressive nature and formal guarantees of LTL to generate provably correct controllers for complex robotic systems. The use of belief space planning for dynamic obstacle belief tracking and heterogeneous robot capabilities to assist one another when environment assumptions are violated allows the planning framework to reduce the conservativeness traditionally associated with using formal methods for robot planning.