Search papers, labs, and topics across Lattice.
1
0
2
This work presents CAST (Critic with Alternating State-value Target), which uses planner-guided behavior to improve value learning while regularizing the value estimate with the current policy.