NEAR trains a noise-conditioned energy-based reward from expert state transitions and anneals the noise level during RL to imitate humanoid motions at AMP-comparable performance.
The agent aims to reach the goal position at the bottom right (the episode ends when the agent’s position is within some threshold of the goal)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Noise-conditioned Energy-based Annealed Rewards (NEAR): A Generative Framework for Imitation Learning from Observation
NEAR trains a noise-conditioned energy-based reward from expert state transitions and anneals the noise level during RL to imitate humanoid motions at AMP-comparable performance.