Multi-Task Reinforcement Learning of Drone Aerobatics by Exploiting Geometric Symmetries
Zhanyu Guo,, Zikang Yin,, Guobin Zhu,, Shiliang Guo, and Shiyu Zhao∗
AI summary
Problem
Conventional reinforcement learning methods struggle with low data efficiency and poor generalization when training a single policy to master multiple aggressive drone maneuvers without relying on manually designed waypoint sequences.
Approach
The authors propose GEAR, a unified multi-task reinforcement learning framework that embeds the inherent rotational symmetry of drone physics directly into its neural network architecture. By combining a symmetry-aware policy backbone with flexible task-specific modulation and separate value estimators, the model efficiently learns to execute diverse maneuvers from a single policy.
Key results
- Achieves 98.85% success rate across diverse aerobatic tasks in simulation
- Outperforms baseline RL methods by 9.53% in final training return
- Successfully deploys a single unified policy on physical MAVs for real-world maneuvers
- Enables composition of basic flight primitives to execute complex aerobatics like Power Loops and Multi-Flips
Why it matters
This approach provides a scalable, data-efficient pathway for training agile, multi-maneuver drone controllers, advancing autonomous aerial robotics for applications like racing, search-and-rescue, and freestyle flight.
Abstract
Flight control for autonomous micro aerial vehi- cles (MAVs) is evolving from steady flight near equilibrium points toward more aggressive aerobatic maneuvers, such as flips, rolls, and Power Loop. Although reinforcement learning (RL) has shown great potential in these tasks, conventional RL methods often suffer from low data efficiency and limited generalization. This challenge becomes more pronounced in multi-task scenarios where a single policy is required to master multiple maneuvers. In this paper, we propose a novel end-to- end multi-task reinforcement learning framework, called GEAR (Geometric Equivariant Aerobatics Reinforcement), which fully exploits the inherent SO(2) rotational symmetry in MAV dynamics and explicitly incorporates this property into the policy network architecture. By integrating an equivariant actor network, FiLM-based task modulation, and a multi-head critic, GEAR achieves both efficiency and flexibility in learning diverse aerobatic maneuvers, enabling a data-efficient, robust, and unified framework for aerobatic control. GEAR attains a 98.85% success rate across various aerobatic tasks, significantly outperforming baseline methods. In real-world experiments, GEAR demonstrates stable execution of multiple maneuvers and the capability to combine basic motion primitives to complete complex aerobatics.