Embodied Intelligence Meets Foundation Models: Tactile Sensing and Soft Manipulation

Overview

Adaptability plays a central role for robotic systems to operate effectively in human-centric environments, with dexterous manipulation among its critical focuses. The workshop addresses this topic via three independent yet complementary domains:

(a) Embodied intelligence, where part of the robot’s adaptability is encoded within the robot’s biomechanically inspired morphology; 

(b) Multimodal and tactile sensing, where adaptability stems from the world Physics represented by diverse sensorial streams like vision, tactile, proprioception, robot actions; and 

(c) Foundation models, where adaptability emerges from learning to understand and reason about the world through large and diverse human-centered data, enabling agents to infer intent and plan context-aware actions. 

What to expect?

The first part of the workshop will feature expert talks, highlighting their contributions within these domains. The second part will set the stage for students to present flashtalks, demos, posters, or reproducible GitHub repositories. The main intent of this workshop is to initiate a cross-domain dialogue while acknowledging their individual challenges and potential ways forward, to be held via panel discussion. The ambitious expected outcome is to establish an online channel for organizers, invited speakers, and participants to get together and discuss progress, spearheading towards general-purpose object manipulation for human-centric environments.