RRT-CoLearn: Towards Kinodynamic Planning Without Numerical Trajectory Optimization

Sampling-based kinodynamic planners, such as rapidly-exploring random trees (RRTs), pose two fundamental challenges: computing a reliable (pseudo-)metric for the distance between two random nodes, and computing a steering input to connect the nodes. The core of these challenges is a two point boundary value problem, known to be NP-hard. Recently, the distance metric has been approximated using supervised learning, reducing computation time drastically. Previous work on such learning RRTs use direct optimal control to generate the data for supervised learning. This letter proposes to use indirect optimal control instead, because it provides two benefits: it reduces the computational effort to generate the data, and it provides a low-dimensional parametrization of the action space. The latter allows us to learn both the distance metric and the steering input. This eliminates the need for a local planner in learning RRTs. Experimental results on a pendulum swing up show tenfold speed-up in both the offline data generation and the online planning time.