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.
CLeAR Lab
Collaborative, Learning, and Adaptive Robots Lab at NUS.
We develop Trustworthy Robots.
Abstracting Robot Manipulation Skills via Mixture-of-Experts Diffusion Policies
Skill Mixture-of-Experts Policy (SMP)—a diffusion mixture-of-experts policy that learns a compact orthogonal skill basis and uses sticky routing for scalable, efficient multi-task manipulation.
Octopi: Object Property Reasoning with Large Tactile-Language Models
We introduce PhysiCLeaR, an annotated dataset of everyday objects and tactile readings collected from a Gelsight Mini sensor, as well as Octopi, a system that leverages both tactile representation learning and large vision-language models to perform physical reasoning and inference, given tactile videos of multiple objects.
Don’t Start from Scratch: Behavioral Refinement via Interpolant-based Policy Diffusion
We develop interpolant policies that leverage informative source distributions for imitation learning.
Latest News
More news
Introducing Dr. Eugene Lim!
Eugene Lim has graduated and is now Dr. Lim. Congratulations Eugene! He will be joining Mathrix as a Quant Analyst.
CLeAR wins REAL-I 2026 Embodied AI Challenge!
CLeAR took Overall Champion at REAL-I 2026, the 1st Real-World Embodied AI Learning Challenge, held at ICRA 2026.
On-Manifold Guidance @ ICML'26
We are excited to share that a paper from CLeAR has been accepted to ICML 2026! Here’s a snapshot: Conflict-Aware Additive Guidance for Flow Models under Co...
Steering & Skill Reuse @ R:SS'26
We are excited to share that two papers from CLeAR were accepted to R:SS 2026! More information about the papers is coming soon, but here’s a snapshot: ...
Latest Papers
See all papers
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.
TOPO-Bench: Benchmarking Topological Mapping under Perceptual Aliasing
An open-source framework for evaluating topological mapping, with the first quantitative measure of dataset ambiguity (perceptual aliasing).
Action Hallucination in Generative Vision-Language-Action Models
A theoretical analysis of why generative VLAs produce physically infeasible actions, identifying topological, precision, and horizon barriers that impose unavoidable tradeoffs.
