Embodiment-conditioned Generalist Control for Multirotor Aerial Robots

2026-06-09

AI Summary

This work addresses the challenge of generalizing a single control policy to arbitrary multirotor configurations—including quadrotors, hexacopters, and non-planar or asymmetric designs—for precise position control. The authors propose conditioning a reinforcement learning policy on a physics-informed airframe descriptor: a mass- and inertia-normalized control allocation matrix. Training a compact neural network across a broad distribution of vehicle configurations enables zero-shot transfer to unseen multirotor geometries. Remarkably, the approach achieves successful real-world deployment on physical platforms using only a single policy network. Leveraging the PPO algorithm and a custom NVIDIA Warp-based dynamics simulator, training requires just five minutes on an RTX 3090 GPU and demonstrates robust performance across three significantly different hexacopter hardware platforms.

Abstract

We present a generalist position control policy capable of controlling arbitrary multirotor configurations of a certain rotor count (e.g., hexarotors or quadrotors) with a single set of network weights. The policy is conditioned on a physics-grounded embodiment descriptor: a mass and inertia-normalized control allocation matrix that captures how mass-normalized motor thrusts generate linear and angular accelerations in the body-frame. To train the policy, we sample from a broad distribution of arbitrary multirotor configurations, including non-planar and asymmetric systems, and optimize a single, compact network using Proximal Policy Optimization. Training requires only five minutes on an RTX 3090 GPU using a custom NVIDIA Warp-based dynamics simulator. Through extensive simulation experiments, we show that embodiment conditioning enables robust generalist control across arbitrary morphologies. We demonstrate zero-shot real-world transfer of this generalist policy on three diverse hexarotor systems, including a planar robot, a partially symmetric non-planar system, and a random asymmetric, non-planar configuration.

Problem

• generalist control
• multirotor aerial robots
• embodiment conditioning
• arbitrary morphology
• zero-shot transfer

Innovation

• embodiment-conditioned control
• generalist policy
• multirotor aerial robots
• zero-shot transfer
• control allocation matrix