Not All Layers Need Tuning: Selective Layer Restoration Recovers Diversity
A training-free method that restores selected layers of a post-trained LLM to their pre-trained weights, recovering output diversity while preserving output quality.
Posts
Laruso blog features productivity, tips, inspiration and strategies for massive profits. Find out how to set up a successful blog or how to make yours even better!
APPL: Agent Priors-guided Policy Learning
An agentic robot learning framework connecting policy learning with agent-based skill selection and composition through structural priors.
Introducing Dr. Eugene Lim!
Eugene Lim has graduated and is now Dr. Lim. Congratulations Eugene! He will be joining Mathrix as a Quant Analyst.
Guided Streaming Stochastic Interpolant Policy
A principled inference-time guidance framework for streaming generative robot policies, enabling fast, reactive obstacle avoidance within the action chunk.
SkillVLA: Tackling Combinatorial Diversity in Dual-Arm Manipulation via Skill Reuse
A dual-arm manipulation framework that enables skill reuse—recomposing learned single-arm skills into novel left–right pairings to tackle combinatorial diversity.
CAR Guidance: Staying On-Manifold under Compositional Rewards
We introduce a plug-and-play module that corrects off-manifold drift when guiding flow models with multiple rewards at inference time.
Heterogeneous Tactile Transformer (HTT)
A self-supervised tactile backbone that learns shared representations across heterogeneous tactile sensors, boosting perception and contact-rich manipulation—even with sensors unseen during pretraining.
