PAC-ACT: Post-training Actor-Critic for Action Chunking Transformers
PAC-ACT is a reinforcement-learning post-training framework for pretrained Action Chunking Transformer policies, designed to improve reliability in precision industrial contact manipulation under pose perturbations and contact-force constraints. It addresses distribution shift in contact-rich tasks that behavior-cloned policies suffer from.
Entities
Related
FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space
FlowDAgger is a new method for adapting frozen generative robot policies using human interventions in latent space. It enables sample- and compute-efficient correction of failure modes outside the pretraining distribution without large-scale data collection or online RL.
CLAP paper proposes direct VLM-to-VLA adaptation to isolate semantic transfer
A new arXiv paper introduces CLAP, a method that converts pretrained vision-language models (VLMs) into vision-language-action models (VLAs) with minimal architectural changes, avoiding the distribution mismatch caused by predicting actions as bare numeric tokens. The approach aims to clarify how VLM capabilities transfer across model scales for robot control.
LLM/VLA models enable prompt-driven exploration in RL
A new research paper proposes using large language models (LLMs) and vision-language-action (VLA) models to drive exploration in reinforcement learning by modifying natural language prompts, which induce global behavioral changes beyond standard action noise. This approach could help policies escape weak local optima more effectively.
KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling
A new paper introduces KV-PRM, a process reward model that reuses KV-cache from the base model to avoid re-encoding entire trajectories, reducing scoring cost from quadratic to linear in sequence length. This enables efficient test-time scaling for long-context multi-agent systems.
Mach-Mind-4-Flash Technical Report Released: 35B MoE Model Matches 100B-Class Performance via Post-Training
A technical report on arxiv introduces Mach-Mind-4-Flash, a 35B-parameter Mixture-of-Experts model with 3B activated parameters. Through post-training optimization alone, it achieves performance on par with or surpassing 100B-parameter-class models, enabled by scalable agentic interaction environments for large-scale reinforcement learning.