The Latent That Never Was: A Forensic Re-run of the CVAE Ablation in Action Chunking Transformer

2026-09-15

AI Summary

研究重新运行了ACT中CVAE编码器的消融实验，发现原论文报告的成功率下降未再现，探讨了训练长度和检查点选择对结果的影响。

Abstract

Action Chunking Transformers (ACT) are widely used to learn robot manipulation from demonstrations. Their conditional variational autoencoder includes an encoder meant to capture differences between demonstrations during training. The original ACT paper reported that encoder removal dropped the mean success rate from 35% to 2% on two simulated tasks with human demonstrations. We re-ran this ablation in the original code and checked whether the findings depend on the implementation or training data. The published drop does not reappear in our tests, although smaller gains or losses in success rate remain uncertain. To investigate the discrepancy, we varied training length and how checkpoints are selected for evaluation. Both can reverse which policy scores higher, but the published drop's cause remains unknown. Success rates alone leave open whether the encoder provides information that helps the policy reconstruct demonstrated actions. On the tested ACT benchmark, the sampled latent provides little reconstruction benefit at every tested nonzero weight of the penalty on latent information. At inference, ACT leaves this latent unused and sets it to zero. Skipping the encoder increases training throughput in both implementations we timed. We release code, evaluation tools and results so others can repeat the comparisons and test the encoder on other tasks.

Problem

• Action Chunking Transformers
• encoder
• success rate
• reconstruction

Innovation

• Action Chunking Transformers
• CVAE Ablation
• Latent Information
• Training Throughput
• Policy Reconstruction