Text-to-motion reinforcement learning
An in-progress simulator exploring language-conditioned motion for a six-axis arm.
Developing the simulator, policy experiments and PyTorch-to-Unity inference path.
In progress
- Exploring language-conditioned motion for a six-axis arm in simulation.
- Developing policy experiments and a PyTorch-to-Unity inference path.
- Result / current state
- The current work is an exploratory research system without a validated headline metric.
- Technologies
- Python
- PyTorch
- Unity
What it does
The project explores turning a natural-language instruction into joint-space motion for a six-axis arm in a physics simulator.
My contribution
I am developing the simulator, goal-conditioned policy experiments and PyTorch-to-Unity closed-loop inference path.
Technical highlight
Language embeddings condition reinforcement-learning policies rather than directly emitting unbounded motor commands; the Unity loop provides the simulated control environment.
Results and status
This is in progress. It is a research prototype without a validated real-arm result or headline performance claim.