MAVEN: A Meta-Reinforcement Learning Framework for Varying-Dynamics Expertise in Agile Quadrotor Maneuvers

2026-03-11

AI Summary

This work proposes the MAVEN framework to address the limited generalization of reinforcement learning policies for quadrotors under significant dynamic changes, such as abrupt mass variations or substantial single-motor thrust loss. By integrating meta-reinforcement learning with a novel predictive context encoder, MAVEN enables a single policy to perform end-to-end agile control across diverse dynamics through online inference of system properties from interaction history. The approach demonstrates, for the first time on a real quadrotor, strong zero-shot sim-to-real transfer with high adaptability and maneuverability. Experimental results show stable high-speed flight under extreme conditions—including up to 66.7% mass change or 70% thrust loss in a single motor—with policy training converging in under one hour.

Technology Category

• Machine Learning: Reinforcement Learning
• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Reinforcement learning (RL) has emerged as a powerful paradigm for achieving online agile navigation with quadrotors. Despite this success, policies trained via standard RL typically fail to generalize across significant dynamic variations, exhibiting a critical lack of adaptability. This work introduces MAVEN, a meta-RL framework that enables a single policy to achieve robust end-to-end navigation across a wide range of quadrotor dynamics. Our approach features a novel predictive context encoder, which learns to infer a latent representation of the system dynamics from interaction history. We demonstrate our method in agile waypoint traversal tasks under two challenging scenarios: large variations in quadrotor mass and severe single-rotor thrust loss. We leverage a GPU-vectorized simulator to distribute tasks across thousands of parallel environments, overcoming the long training times of meta-RL to converge in less than an hour. Through extensive experiments in both simulation and the real world, we validate that MAVEN achieves superior adaptation and agility. The policy successfully executes zero-shot sim-to-real transfer, demonstrating robust online adaptation by performing high-speed maneuvers despite mass variations of up to 66.7% and single-rotor thrust losses as severe as 70%.

Problem

• quadrotor dynamics
• generalization
• adaptability
• agile navigation
• dynamic variations

Innovation

• meta-reinforcement learning
• quadrotor agility
• predictive context encoder
• zero-shot sim-to-real transfer
• dynamics adaptation