A2C PandaReachDense-v3

Trained from scratch with four environments and observation/reward normalization. Independent 100-episode evaluation: -0.21 +/- 0.10; score -0.30.

Load the included vec_normalize.pkl with VecNormalize.load; set training=False and norm_reward=False for evaluation, then load a2c-PandaReachDense-v3.zip with A2C.load.