Dynamically Constrained Motion Planning Networks for Non-Holonomic Robots

Reliable real-time planning for robots is essential in today's rapidly\nexpanding automated ecosystem. In such environments, traditional methods that\nplan by relaxing constraints become unreliable or slow-down for kinematically\nconstrained robots. This paper describes the algorithm Dynamic Motion Planning\nNetworks (Dynamic MPNet), an extension to Motion Planning Networks, for\nnon-holonomic robots that address the challenge of real-time motion planning\nusing a neural planning approach. We propose modifications to the training and\nplanning networks that make it possible for real-time planning while improving\nthe data efficiency of training and trained models' generalizability. We\nevaluate our model in simulation for planning tasks for a non-holonomic robot.\nWe also demonstrate experimental results for an indoor navigation task using a\nDubins car.\n