Honda Research Institute Europe
Choose a step to see what happens.
Start in the real world
The real scene and its two camera views define the task we want to learn.
“Pick up the vial and place it on the rack.”
Reads the recorded views, identifies the scene and helps turn the task into an experiment.
Real → simulation
Translate the observed workspace into a MuJoCo scene, then inspect the robot and task in motion.
Builds the scene, sets up robot geometry and cameras, and checks simulated task motion.
Simulation → appearance
Keep the motion. Change the materials, lighting and camera appearance.
Drag the divider to compare both cameras.
Builds materials, lighting and camera views, then renders saved simulated trajectories.
A place to develop behavior
Try executable control code and a policy that learns from reward.
“Pick up the vial and place it on the rack.”
Camera images and robot state. A policy trained from scratch.
Writes and tests the task controller, inspects the trajectory and prepares rendered observations.
Data → adaptation and prediction
Adapt a foundation policy and explore a prototype for predicting future motion.
Explore fine-tuning with rendered data and with real demonstrations.
Prepares the datasets, configures π0.5 fine-tuning and reviews the resulting behavior.
Real observations