Imitation Learning-Based Online Time-Optimal Control with Multiple-Waypoint Constraints for Quadrotors

TL;DR

The paper tackles online time-optimal quadrotor trajectories through multiple waypoints in confined spaces by learning a control policy from CPC-generated data. It introduces WN&CNets (waypoint-constrained navigation and control networks) trained via imitation learning and augments them with a MINCO-based transition phase to bridge waypoint switches without hover. Key contributions include a CPC-derived dataset for two-waypoint trajectories, a neural network architecture with online prediction capability, and a transition strategy enabling multi-waypoint flights, validated in simulation and real flights with up to seven waypoints. The results demonstrate near-time-optimal online performance with significantly reduced computation time, making aggressive waypoint-constrained flight feasible in dynamic or cluttered environments.

Abstract

Over the past decade, there has been a remarkable surge in utilizing quadrotors for various purposes due to their simple structure and aggressive maneuverability, such as search and rescue, delivery and autonomous drone racing, etc. One of the key challenges preventing quadrotors from being widely used in these scenarios is online waypoint-constrained time-optimal trajectory generation and control technique. This letter proposes an imitation learning-based online solution to efficiently navigate the quadrotor through multiple waypoints with time-optimal performance. The neural networks (WN&CNets) are trained to learn the control law from the dataset generated by the time-consuming CPC algorithm and then deployed to generate the optimal control commands online to guide the quadrotors. To address the challenge of limited training data and the hover maneuver at the final waypoint, we propose a transition phase strategy that utilizes MINCO trajectories to help the quadrotor 'jump over' the stop-and-go maneuver when switching waypoints. Our method is demonstrated in both simulation and real-world experiments, achieving a maximum speed of 5.6m/s while navigating through 7 waypoints in a confined space of 5.5m*5.5m*2.0m. The results show that with a slight loss in optimality, the WN&CNets significantly reduce the processing time and enable online optimal control for multiple-waypoint constrained flight tasks.

Figures (13)

• Figure 1: The quadrotor continuously passes through moving target points at high agility with the proposed WN&CNets and the MINCO transition.
• Figure 2: Datasets that contain a great amount of state-control pairs are generated using the CPC method. These state-control pairs are then taken as training data for WN&CNets. The WN&CNets will then be deployed to guide the quadrotor to fly through multiple waypoints.
• Figure 3: Illustrative sketch of the CPC and its deficiency in multi-drone racing scenarios.
• Figure 4: The sketch of the waypoint-switch strategy: with the proposed transition phase, the quadrotors don't have to hover at the waypoints.
• Figure 5: The quadrotor vehicle.