Improving Incremental Planning Performance through Overlapping Replanning and Execution

Abstract

Deployment of motion planning algorithms in practical applications has lagged due to their slow speed in reacting to disturbances. We believe that the best way to address this is to reuse learned planning and control information across queries. In previous work, we introduced Chekov, a reactive, integrated motion planning and execution system that reuses learned information in the form of an enhanced roadmap. We have previously shown how we can use Chekov to formulate trajectory optimization problems that result in superior performance in static environments. In this work, we show how incremental planning can be incorporated into the formulation of optimized trajectories from roadmap seed trajectories. Further, we show how an incremental planner can be adapted to reduce the overhead incurred for replanning when trajectories become invalid during execution.

Fast-Reactive Probabilistic Motion Planning for High-Dimensional Robots

“…In this way, p-Chekov will not need to waste a long time trying to search for feasible solutions for infeasible cases. Another potential future work direction is to incorporate online obstacle avoidance [36,49,55] into p-Chekov, so that it can handle dynamic obstacles in the execution environment. Additionally, real robot experiments with raw sensor data are also necessary before p-Chekov can be deployed in real-world applications.…”

“…Dynamic obstacles could be handled through storing redundant roadmap paths and by coupling these paths with fast online obstacle-aware optimization. In addition, the incremental Chekov approach introduced by [49] is an extension to the deterministic Chekov approach illustrated in this paper that highlights the handling of dynamic obstacles in the environment. Incorporating incremental Chekov's ability of handling dynamic obstacles into the chance-constrained motion planning framework described in this paper is a potential direction of our future research.…”

“…To address these difficulties, we propose probabilistic Chekov (p-Chekov), a combined sampling-based and optimization-based approach that takes advantage of the fact that most obstacles in a lot of practical motion planning tasks are static and only a small number of objects are dynamic during deployment. In these cases, we can construct sparse roadmaps based on our prior knowledge about the static environment to cache feasible trajectories offline, so that during plan execution, we only need to optimize solution trajectories according to new observations [19,49] and adjust plans to satisfy safety requirements [20]. Combining ideas from risk allocation [46,47] and supervised learning, p-Chekov can effectively reason over uncertainties and provide motion plans that satisfy constraints over the probability of plan failure, i.e., chance constraints [47].…”

Chance Constrained Motion Planning for High-Dimensional Robots

“…Dynamic obstacles could be handled through storing redundant roadmap paths and by coupling these paths with fast online obstacle-aware optimization. In addition, the incremental Chekov approach introduced by Orton et al (2019) is an extension to the deterministic Chekov approach illustrated in this paper that highlights the handling of dynamic obstacles in the environment. Incorporating incremental Chekov's ability of handling dynamic obstacles into the chance-constrained motion planning framework described in this paper is a potential direction of our future research.…”

“…In order to address these difficulties, we propose probabilistic Chekov (p-Chekov), a combined sampling-based and optimization-based approach that takes advantage of the fact that most obstacles in a lot of practical motion planning tasks are static and only a small number of objects are dynamic during deployment. In these cases, we can construct sparse roadmaps based on our prior knowledge about the static environment to cache feasible trajectories offline, so that during plan execution, we only need to optimize solution trajectories according to new observations (Dai et al 2018;Orton et al 2019) and adjust plans to satisfy safety requirements (Dai et al 2019). Combining ideas from risk allocation (Ono and Williams 2008a,b) and supervised learning, p-Chekov can effectively reason over uncertainties and provide motion plans that satisfy constraints over the probability of plan failure, i.e.…”