Entropy Coefficient
A coefficient for a regularization term in the reinforcement learning objective function that controls the stochasticity of the policy to balance exploration and exploitation and avoid premature convergence to local optima.
Timeline (last 90 days)
Training introduced mechanisms such as penalizing the torque distribution to encourage the four legs to exert force evenly, penalizing foot slipping, applying a low-pass filter to actions, and increasing the entropy coefficient to enhance exploration after the curriculum schedule reached its upper limit.
Increasing the entropy coefficient after curriculum learning reaches its upper limit is a "exploit first, then explore" strategy: low entropy in the early stage quickly masters the basic gait, while high entropy in the later stage helps the policy find more robust solutions within the disturbance range it has adapted to.
All Facts (2)
Training introduced mechanisms such as penalizing the torque distribution to encourage the four legs to exert force evenly, penalizing foot slipping, applying a low-pass filter to actions, and increasing the entropy coefficient to enhance exploration after the curriculum schedule reached its upper limit.
50%UnverifiedIncreasing the entropy coefficient after curriculum learning reaches its upper limit is a "exploit first, then explore" strategy: low entropy in the early stage quickly masters the basic gait, while high entropy in the later stage helps the policy find more robust solutions within the disturbance range it has adapted to.
50%