Learning State-Dependent Losses for Inverse Dynamics Learning

Being able to quickly adapt to changes in dynamics is paramount in model-based control for object manipulation tasks. In order to influence fast adaptation of the inverse dynamics model's parameters, data efficiency is crucial. Given observed data, a key element to how an optimizer updates model parameters is the loss function. In this work, we propose to apply meta-learning to learn structured, state-dependent loss functions during a meta-training phase. We then replace standard losses with our learned losses during online adaptation tasks. We evaluate our proposed approach on inverse dynamics learning tasks, both in simulation and on real hardware data. In both settings, the structured and state-dependent learned losses improve online adaptation speed, when compared to standard, state-independent loss functions.

Authors

• Kristen Morse
• Neha Das
• Yixin Lin 0003
• Austin S. Wang
• Akshara Rai
• Franziska Meier

Keywords

• Training
• Adaptation models
• Analytical models
• Data models
• Hardware
• Task analysis
• Standards
• Inverse Dynamics
• Loss Function
• Dynamic Changes
• Dynamic Model
• Adaptive Model
• Inverse Model
• Model-based Control
• Real Hardware
• Model Analysis
• Neural Network
• Time Step
• Learning Rate
• Simulated Data
• Simulation Experiments
• Stochastic Gradient Descent
• Online Learning
• Cognitive Map
• Data Streams
• Loss Of Structure
• Torque Values
• Learning Loss
• Mean Square Error Loss
• Static Friction
• Real Robot
• Hardware Experiments
• Adaptive Optimization