TAMPC: A Controller for Escaping Traps in Novel Environments
- Trap-Aware Model Predictive Control
Learning When to Trust a Dynamics Model for Planning in Reduced State Spaces (2020)
Simultaneous learning of hierarchy and primitives for complex robot tasks (2018)

Relevant

Keep it Simple: Data-efficient Learning for Controlling Complex Systems with Simple Models (2021)
- perception system φ that maps images to simple model states (even if the image is generated from the complex system)
- estimate how well a simple model state approximates the complex system at a given state.

Prediction of human motion

Unsupervised Early Prediction of Human Reaching for Human-robot Collaboration in Shared Workspaces (2018)
- Maintain a two-layer library: Layer 1 - GMMs for palm (position) motion and Layer 2 - GMMs for human arm (joint center position) motion.
- For new user, either update an existing GMM or create a new GMM depending on some threshold, in layer 1. Then depending on the GMM in layer 1, we update the corresponding (or create a new) GMMs in layer 2.
- What is a membership-proportional prior?
Goal Set Inverse Optimal Control and Iterative Re-planning for Predicting Human Reaching Motions in Shared Workspaces (2016)
- Collect demonstrations from a pair of users performing a table-top manipulation task.
- Use Path Integral Inverse Reinforcement Learning (PI IRL) to learn the user's cost function i.e. weights for feature functions. Assuption: user's objective while reaching is to minimize this cost function.
- Cost function used to iteratively compute paths to a goal region (not just a single goal configuration as it may be unknown) from the user's current state.

Task Space Regions: A framework for pose-constrained manipulation planning (2011)