Accelerated Low-power Motion Planning for Real-time Interactive Autonomy

The DOD and robotics industry, have a critical need for a motion planning capability that is practical for real-world application of autonomous systems that perform physical motions. Current motion planning systems are too slow; they assume that the environment will be static during motion planning and execution, each of which can take on the order of seconds. Further, these systems typically generate kinematic trajectories, but do not take into account mechanism dynamics, which is needed for fast, agile movement. To address this problem, we propose RAPID, a motion planning system that features four key innovations: First, we will use a Roadmap-based approach that inherently represents the static collision-free trajectory space, and that uses off-line computation to facilitate fast runtime planning. Second, we will cache swept volume representations of motions with the edges of the Roadmap, and use these to implement hardware-accelerated collision detection. Third, we will cache dynamic information in the form of Flow-tubes, and also associated these with edges of the Roadmap, allowing for fast, agile motions. Finally, we will use trajectory optimization algorithms to fine-tune Roadmap trajectories. RAPID will enable robots that collaborate closely with humans, and that operate in unstructured, dynamic, and uncertain environments.