All four of our submissions (3 papers, 1 demo) were accepted to R:SS 2024! A fantastic accomplishment by our CLeAR members Kaiqi, Samson, Tasbolat, Linh, Kelvin, and Eugene. More information about the papers are coming soon but here’s a snapshot:
- Don’t start from scratch: Behavioral Refinement via Interpolant-based Policy Diffusion. Led by Kaiqi, this work enables you to use existing policies with data-driven diffusion. Practically, this leads to lower inference times and data requirements! A pre-print is available here.
- Octopi: Object Property Reasoning with Large Tactile-Language Models. We introduce Octopi, a LLM with the sense of touch for better physical understanding! Octopi comes with PhysiCLeAR, an evaluation suite and training dataset for training future touch-enabled LLMs. See the preprint here.
- GRaCE: Balancing Multiple Criteria to Achieve Stable, Safe, and Task Functional Grasps. Led by Tasbolat, we examine robot grasping in situations multiple criteria (e.g., safety, stability, task functionality) may conflict and differ in importance. GRaCE is a probabilistic framework, which employs hierarchical rule-based logic and a rank-preserving utility function, for trading-off different criteria in a principled manner under uncertainty.
- Demonstrating Arena 3.0: Advancing Social Navigation in Collaborative and Highly Dynamic Environments. This demo paper led by Linh Kastner, showcases the new iteration of Arena. Arena 3.0 is a comprehensive software stack for facilitating the development, simulation, and benchmarking of social navigation approaches in collaborative environments.
