Scalable Multicore Motion Planning Using Lock-Free Concurrency.

We present PRRT (Parallel RRT) and PRRT* (Parallel RRT*), sampling-based methods for feasible and optimal motion planning designed for modern multicore CPUs. We parallelize RRT and RRT* such that all threads concurrently build a single motion planning tree. Parallelization in this manner requires that data structures, such as the nearest neighbor search tree and the motion planning tree, are safely shared across multiple threads. Rather than rely on traditional locks which can result in slowdowns due to lock contention, we introduce algorithms based on lock-free concurrency using atomic operations. We further improve scalability by using partition-based sampling (which shrinks each core's working data set to improve cache efficiency) and parallel work-saving (in reducing the number of rewiring steps performed in PRRT*). Because PRRT and PRRT* are CPU-based, they can be directly integrated with existing libraries. We demonstrate that PRRT and PRRT* scale well as core counts increase, in some cases exhibiting superlinear speedup, for scenarios such as the Alpha Puzzle and Cubicles scenarios and the Aldebaran Nao robot performing a 2-handed task.

Deep learning can accelerate grasp-optimized motion planning.

Robots for picking in e-commerce warehouses require rapid computing of efficient and smooth robot arm motions between varying configurations. Recent results integrate grasp analysis with arm motion planning to compute optimal smooth arm motions; however, computation times on the order of tens of seconds dominate motion times. Recent advances in deep learning allow neural networks to quickly compute these motions; however, they lack the precision required to produce kinematically and dynamically feasible motions. While infeasible, the network-computed motions approximate the optimized results. The proposed method warm starts the optimization process by using the approximate motions as a starting point from which the optimizing motion planner refines to an optimized and feasible motion with few iterations. In experiments, the proposed deep learning-based warm-started optimizing motion planner reduces compute and motion time when compared to a sampling-based asymptotically optimal motion planner and an optimizing motion planner. When applied to grasp-optimized motion planning, the results suggest that deep learning can reduce the computation time by two orders of magnitude (300×), from 29 s to 80 ms, making it practical for e-commerce warehouse picking.

Cache-Aware Asymptotically-Optimal Sampling-Based Motion Planning.

We present CARRT* (Cache-Aware Rapidly Exploring Random Tree*), an asymptotically optimal sampling-based motion planner that significantly reduces motion planning computation time by effectively utilizing the cache memory hierarchy of modern central processing units (CPUs). CARRT* can account for the CPU's cache size in a manner that keeps its working dataset in the cache. The motion planner progressively subdivides the robot's configuration space into smaller regions as the number of configuration samples rises. By focusing configuration exploration in a region for periods of time, nearest neighbor searching is accelerated since the working dataset is small enough to fit in the cache. CARRT* also rewires the motion planning graph in a manner that complements the cache-aware subdivision strategy to more quickly refine the motion planning graph toward optimality. We demonstrate the performance benefit of our cache-aware motion planning approach for scenarios involving a point robot as well as the Rethink Robotics Baxter robot.