🤖🍨 Sundae Robotics 03: Scaling Touch, Multimodal Perception & Dexterous Manipulation with Flexible Tactile Skin for Robot Learning

🤖🍨 Grab a sundae and join Sundae Robotics, a private, invite-only Sunday series bringing together robotics researchers, founders, and builders working at the frontier of physical intelligence.

Sundae Robotics 03

Scaling Touch & Multimodal Robot Learning
Featured Talk: Flexible Tactile Skin for Dexterous Manipulation

Keynote: Binghao Huang
Ph.D. Student, Columbia University (Yunzhu Li) · Former NVIDIA Seattle Robotics Lab · M.S., UC San Diego (Xiaolong Wang)

Touch is one of the most information-rich sensing modalities available to robots, yet it remains dramatically underutilized compared to vision. Humans rely on tactile feedback for nearly every contact-rich manipulation task, from inserting connectors and threading cables to grasping fragile objects and assembling mechanical components. For robots to achieve similarly robust dexterity, they must not only sense contact but also learn scalable representations that combine touch with vision and simulation.

In this talk, Binghao will present a unified approach toward scaling tactile intelligence for robot learning. First, he will introduce a low-cost, flexible tactile skin capable of covering large robot surfaces while producing high-quality tactile observations for manipulation. He will discuss the practical design tradeoffs behind tactile hardware, how these sensors are integrated into robotic systems, and why flexible tactile sensing enables capabilities that vision alone cannot provide. He will then present multimodal learning approaches that jointly encode vision and touch for robot decision-making, followed by two complementary strategies for scaling tactile data: collecting large-scale real-world tactile datasets using portable sensing devices and leveraging calibrated GPU-parallel tactile simulation to improve policy robustness through reinforcement learning and real-to-sim-to-real transfer. Together, these approaches demonstrate a practical roadmap toward tactile-enabled robot foundation models capable of robust contact-rich manipulation.

Binghao's research spans robot learning, dexterous manipulation, tactile sensing, multimodal perception, sim-to-real transfer, and reinforcement learning. His recent work focuses on scalable tactile hardware, multimodal robot learning, tactile data collection, and tactile simulation for dexterous manipulation. He is currently a Ph.D. student at Columbia University advised by Yunzhu Li and previously worked at NVIDIA Seattle Robotics Lab.

Pre-Reading

• Scaling Touch: Flexible Tactile Skin for Dexterous Manipulation
• Vision-Touch Multimodal Learning for Robot Manipulation

Topics

• Flexible tactile sensing for dexterous manipulation
• Multimodal robot learning with vision and touch
• Scaling tactile datasets for robot foundation models
• Tactile simulation and real-to-sim-to-real transfer
• Contact-rich manipulation and reinforcement learning

Open Discussion + Q&A

• Will touch become a core modality for future robot foundation models?
• How should tactile data be collected and scaled beyond teleoperation?
• What representations best fuse vision, touch, and proprioception?
• Can tactile simulation meaningfully reduce the need for large-scale real-world data?
• What manipulation capabilities remain fundamentally impossible without tactile sensing?