GPU-Accelerated Policy Optimization via Batch Automatic Differentiation of Gaussian Processes for Real-World Control
Link to Published Article → IEEE 2022 International Conference on Robotics and Automation (ICRA)
Video
Highlights
- Scalable policy optimization algorithm for actuator feedback control
- GPU-acceleration: Rollout, backpropagation, and optimization steps performed all on GPU
- Achieves sub-minute training time per actuator on modern consumer GPUs
-
Machine control with Python over CAN interface (Codes:
CANAutomate)
Presentation at International Conference on Robotics and Automation (ICRA) in Philadelphia, PA, USA
Citation
Abdolreza Taheri, Joni Pajarinen, and Reza Ghabcheloo. “GPU-Accelerated Policy Optimization via Batch Automatic Differentiation of Gaussian Processes for Real-World Control”. In: International Conference on Robotics and Automation (ICRA) (2022), pp. 10557-10563. DOI: 10.1109/ICRA46639.2022.9811876. [© 2022 IEEE].