MPC-MPNet: Model-Predictive Motion Planning Networks for Fast, Near-Optimal Planning Under Kinodynamic Constraints

Kinodynamic Motion Planning (KMP) is to find a robot motion subject concurrent kinematics and dynamics constraints. To date, quite few methods solve KMP problems those that exist struggle near-optimal solutions exhibit high computational complexity as the planning space dimensionality increases. address these challenges, we present scalable, imitation learning-based, Model-Predictive Networks framework quickly finds path with worst-case theoretical guarantees under kinodynamic constraints for practical underactuated systems. Our introduces two algorithms built on neural generator, discriminator, parallelizable Model Predictive Controller (MPC). The generator outputs various informed states towards given target, discriminator selects best possible subset from them extension. MPC locally connects selected while satisfying leading feasible, solutions. We evaluate our range of cluttered, kinodynamically constrained, results indicating significant improvements in computation times, qualities, success rates over existing methods.