Learning Hybrid Policies for MPC with Application to Drone Flight in Unknown Dynamic Environments

This paper addresses the challenge of controlling a drone in unknown dynamic environments using a new model predictive control (MPC) approach called hyMPC. The hyMPC leverages high-level decision variables to adapt to uncertain environmental conditions and uses neural network policies for real-time decision-making. The method was tested in numerical simulations, achieving a 100% success rate in 20 drone flight tests traversing a swinging gate, demonstrating its ability to achieve safe and precise flight with limited prior knowledge of environmental dynamics.

Publication date: 19 Jan 2024
Project Page: N/A
Paper: https://arxiv.org/pdf/2401.09705