---

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

We introduce a novel hypernetwork-based framework for Multi-Task Reinforcement Learning (MTRL) that addresses these limitations by learning a single, generalizable policy for multiple diverse robotic systems:

• High Speed Racing: Our framework enables a single policy to successfully race on a variety of unseen tracks.
• Floating Platform: a single hypernetwork policy effectively performs four distinct control objectives: stabilization, docking, velocity tracking, and rendezvous.
• Sim-to-real: validation for the floating platform tasks.
• Code and Weights: Opens-source of the entire stack.

Getting Started

Clone the repo and go to the corresponding branch:

git clone 
cd Hyper-GNC

Build and start the docker

./docker/container.py build
./docker/container.py start

Train

./scripts/reinforcement_learning/rsl_rl/control_train_hypernet.sh

Eval

You can download the weights form the google drive and add them inside the logs folder of the docker. Download link.

./scripts/reinforcement_learning/rsl_rl/control_eval_hypernet.sh