Tag
The paper proposes RSIQL, a non-hierarchical method for offline goal-conditioned reinforcement learning that uses selective reward stimulation to improve performance on benchmarks like D4RL.