Robots in semi-structured environments such as homes and warehouses, sporadically require computation of high-dimensional motion plans. Cloud and fog-based parallelization of motion planning can speed up planning, but allocating always-on high-end computers for sporadic computations can be less efficient than a new class of “serverless” computing that can be allocated on-demand. This paper proposed parallelizing the computation of a sampling-based multi-query motion planner based on asymptotically-optimal Probabilistic Road Maps (PRM*) using the simultaneous execution of 100sof cloud-based serverless functions. We explore how to over-come the inherent limitations of serverless computing related to communication and bandwidth limitations using different work-sharing techniques and provide proofs of probabilistic completeness and asymptotic optimality. In experiments on synthetic benchmarks and on a physical Fetch robot performing a sequence of decluttering motions, we observe up to a 50x speedup relative to a 4 core setup with only a marginally higher cost.

Experiments are run on real-world decluttering scenarios (with the Fetch robot) and synthetic benchmarks. The first row for each scenario depicts the tradeoff between cost and end-time that a user can make: the size of the dot indicates the number of packets sent. Higher numbers of lambdas always finish quicker but have a marginally higher cost; and the best work-sharing method is scenario dependent. The second row for each scenario shows the time scaling of the algorithm with an increasing number of lambdas: as the number of lambdas increases the startup overhead and sampling time costs dominate the overall computation that causes a reduction in parallel efficiency.