Residual Learning
Residual learning is a hybrid modeling approach that combines physics-based models with data-driven models. A physics model provides baseline predictions, while a neural network learns to identify and correct systematic errors in that model. The approach combines the extrapolation capability of physical knowledge with the adaptive calibration capability of data-driven methods.
Timeline (last 90 days)
ORL adopts on-policy residual learning: training data is generated in real time by the current model to ensure that the training distribution matches the inference distribution.
The combination of physics-based models and data-driven models is referred to in robotics as 'hybrid modeling' or 'residual learning,' with the typical approach being that the physics model provides a baseline prediction while a neural network learns the residual error.
All Facts (2)
ORL adopts on-policy residual learning: training data is generated in real time by the current model to ensure that the training distribution matches the inference distribution.
50%UnverifiedThe combination of physics-based models and data-driven models is referred to in robotics as 'hybrid modeling' or 'residual learning,' with the typical approach being that the physics model provides a baseline prediction while a neural network learns the residual error.
50%