Learning motion plans using Neural ODEs

This is an ongoing project with Dr. Matni and Dr. Figueroa. We learn a dynamical systems based nominal motion plan parameterized by Neural ODEs using 2/3 demonstrations  of a task from kinesthetic teaching.  We enforce stability and safety by solving a quadratic program online at task execution using control theoretic tools. We integrate a high-level temporal logic based planning framework for reactive and interactive tasks. We demonstrate our approach on the Franka Emika robot arm for nonlinear periodic tasks such as wiping and stirring.

Nawaz, F., Li, T., Matni, N., & Figueroa, N. (2023). "Learning complex motion plans using Neural ODEs with safety and stability guarantees". accepted to ICRA 2024. Project webpage

Nawaz, F., Peng, S., Lindemann, L., Figueroa, N., & Matni, N. (2024). "Reactive Temporal Logic-based Planning and Control for Interactive Robotic Tasks". accepted to IROS 2024. Project webpage

Kinesthetic teaching

Wiping a human mannequin

Wiping a white board

Nominal task execution

Reactive obstacle avoidance (safety)

Disturbance rejection (stability)

Stirring task

Kinesthetic teaching

Task execution