Discrete vs. Continuous Latent Actions in CLAM for Robot Imitation Learning

This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the discrete latent action formulation in CLAM compare to continuous approaches in terms of sample efficiency and final success rates on the CALVIN benchmark when trained with varying levels. 8 claims were extracted from source literature; 7 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.0/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does the discrete latent action formulation in CLAM compare to continuous approaches in terms of sample efficiency and final success rates on the CALVIN benchmark when trained with varying levels of demonstration diversity?

Autonomous literature synthesis. Automated review score: 8.0/10. Full text and citation available at Assignee Research.