Vision-Based Deep Reinforcement Learning of UAV-UGV Collaborative Landing Policy Using Automatic Curriculum

Abstract

Collaborative autonomous landing of a quadrotor Unmanned Aerial Vehicle (UAV) on a moving Unmanned Ground Vehicle (UGV) presents challenges due to the need for accurate real-time tracking of the UGV and the adjustment for the landing policy. To address this challenge, we propose a progressive learning framework for generating an optimal landing policy based on vision without the need of communication between the UAV and the UGV. First, we propose the Landing Vision System (LVS) to offer rapid localization and pose estimation of the UGV. Then, we design an Automatic Curriculum Learning (ACL) approach to learn the landing tasks under different conditions of UGV motions and wind interference. Specifically, we introduce a neural network-based difficulty discriminator to schedule the landing tasks according to their levels of difficulty. Our method achieves a higher landing success rate and accuracy compared with the state-of-the-art TD3 reinforcement learning algorithm.

Autonomous Landing of Quadrotor Unmanned Aerial Vehicles Based on Multi-Level Marker and Linear Active Disturbance Reject Control

“…A hemispherical infrared marker was proposed for the UAV autonomous landing on a moving ground vehicle, and autonomous landing experiments were operated to demonstrate the effectiveness from various angles [ 11 ]. An approach based on deep reinforcement learning was designed for the UAV landing on a moving unmanned ground vehicle (UGV), which achieved a high landing success rate and accuracy [ 12 ]. This approach did not have any specific communication between the UAV and UGV.…”

Genetic-Algorithm-Aided Deep Reinforcement Learning for Multi-Agent Drone Delivery

“…Agents in standard reinforcement learning start to learn the environment through random policies to find a policy that is close to the optimal policy for a given task. In the curriculum learning setting, instead of directly learning a difficult task, the agent is exposed to simplified versions of the main problem [60]. In the context of a planning problem, Curriculum Learning could involve starting with a simple environment where the agent has to navigate simple, straightforward routes and deliver packages to fixed locations.…”