Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
University of Ottawa, Vector Institute
2
0
4
Sample efficiency isn't everything—PPO outperforms SAC and TD-MPC2 in training speed, challenging conventional wisdom in RL algorithm selection.
Choosing the right action space can make or break the performance of vision-based robotic manipulation, with joint velocity proving to be the standout option.