RealFeedApp ResMerge reduces number of stored parameters
The ResMerge method uses a pre-trained policy to learn corrections via residual learning, requiring no additional demonstrations for new tasks. It consolidates compatible residuals in parameter space to reduce stored data volume.
- Field
- robotics
- What they did
- Developed the ResMerge method, which allows robots to adapt to new tasks without needing to collect new action demonstrations, using corrective parameters based on prior experience.
- Why it matters
- This simplifies robot adaptation in the real world by eliminating the need for continuous data collection and model parameter growth.
#robotics#continual learning#residual policy#reinforcement learning
