Learning Adaptive Multi-Task Guidance, Navigation, and Control via Hypernetworks

2026-07-27

AI Summary

This work addresses the limitations of conventional orbital robotic systems, which rely on task-specific controllers that are fragile and poorly adaptable to evolving mission requirements. To overcome this, the authors propose the HYPER-GNC framework, which employs a hypernetwork to map physics-informed task embeddings onto the weights of a shared policy network, thereby constructing a continuous task embedding space. This enables a single, compact controller to simultaneously master four distinct GNC tasks—velocity tracking, docking, inspection, and obstacle avoidance—and generalize to novel task configurations without retraining. The approach matches the sample efficiency of single-task expert models while demonstrating strong robustness against inertial uncertainties and external disturbances. Notably, it achieves the first successful sim-to-real transfer of multitask GNC capabilities, validated on a physical satellite simulator.

Abstract

Autonomous free-flying robots in orbital environments require controllers that are both versatile and resource-efficient, yet maintaining a separate, task-specific policy for each mission profile is architecturally brittle and limits operational flexibility as requirements evolve. We introduce HYPER-GNC, a multi-task reinforcement learning framework in which a hypernetwork maps physics-informed task embeddings to the weights of a shared actor-critic policy, enabling a single compact controller to master four distinct GNC tasks: velocity tracking, docking, inspection, and navigation with obstacle avoidance. The continuous embedding space allows the controller to generalize to novel mission configurations at deployment time without any retraining. Extensive experiments demonstrate that HYPER-GNC achieves sample efficiency comparable to single-task specialists while maintaining stability under significant inertial perturbations and external body wrenches. We further validate the framework on a physical satellite emulator, successfully bridging the simulation-to-reality gap across all mission profiles. Code, trained models, and deployment scripts are made publicly available to support reproducibility.

Problem

• multi-task control
• autonomous robots
• orbital environments
• task-specific policy
• operational flexibility

Innovation

• hypernetworks
• multi-task reinforcement learning
• adaptive control
• GNC
• sim-to-real transfer