VALT computes the optimal observation adversary from the agent's own value function via soft-constrained optimization, enabling sample-efficient off-policy robust RL without a separately trained attacker.
We denote the ablation setting as w/o PE, and regularization is applied in all settings, including the ablation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation
VALT computes the optimal observation adversary from the agent's own value function via soft-constrained optimization, enabling sample-efficient off-policy robust RL without a separately trained attacker.