Intelligence, in contact.
The contact layer of intelligent machines.
Explore source-backed robotics research across robot skin, tactile sensors, robot hands, and Physical AI. Find the papers. Understand the evidence.
Touch study 01 AI-generated concept · not an experimental image
01 / Research dispatch
ActiveWAM treats where a robot looks as part of the same action sequence as bimanual manipulation, while using transformed video histories only during training.
A tank-mounted wearable robot uses two underwater thrusters as directional haptic cues, but its wall-following behavior depends on a distance estimate that failed in one documented run.
HumanoidToolBench tracks contact, correct selection, lifting and final success across stationary and mobile Unitree G1 tool-use tasks instead of collapsing them into one score.
02 / Explore the field
Robot Skin → Tactile AI → Physical AI
Choose a starting point. Follow the evidence.
Robot skin detects contact, pressure, shear and slip across hands, grippers and body surfaces. It gives robots information about physical interaction that vision alone cannot provide.
Start with the fundamentalsRoboSkin Working Papers
All working papers ↗Research in progress / Version 0.2
When should a robot probe before acting? Explore our research proposal and its synthetic decision benchmark.
Not peer reviewed · No physical-robot evaluation
Explore the proposal03 / Research desk
Start with a Python tactile-data exercise. Then explore ROS 2 messages, calibration and replay.
Artificial intelligence supplies perception, prediction, reasoning, and action policies. Robotics supplies the sensors, controllers and physical feedback.
Open the AI and robotics field guide ↗Language defines the task while vision, robot state, and touch describe the current physical situation.
Open layer →Models turn images, joint state, force, and contact signals into representations the robot can use.
Open layer →A reasoning layer estimates state, plans task steps, and decides which capability should act next.
Open layer →An action model converts observations and instructions into trajectories, poses, or motor commands.
Open layer →Controllers and actuators turn a proposed action into timed movement on a specific robot body.
Open layer →Hands, grippers, feet, or body surfaces meet real objects, people, tools, and changing environments.
Open layer →Tactile AI detects contact, slip, force, and deformation so control or learning can respond to what actually happened.
Open layer →Physical contact returns evidence to perception, control, and learning. This is the layer where robot skin and tactile AI make a general robot model contact-aware.
Follow the learning loop →In the RoboSkin context, Physical AI means physical-world AI systems that use sensing, reasoning and control to act. Touch connects those actions to contact, pressure and slip.
Track humanoid robots, Physical AI, embodied AI, and robot manipulation
The September 17 preprint adds human tactile demonstrations to robot world-model training. With robot pretraining data held fixed, its largest human-data condition improves reported contact prediction, while synthetic policy-training data produces mixed real-robot outcomes.
Our review separates prediction quality, policy evaluation, and downstream robot performance, with practical implications for recording touch, preserving data provenance, and evaluating generated trajectories.
Reviewed 2026-09-18 · Source date 2026-09-17Robot skin is a tactile sensing surface that helps robots detect contact, pressure, shear, slip, and interaction events across hands, grippers, arms, or curved body surfaces. It gives Physical AI systems a contact layer that vision alone cannot provide.
Open the robot skin glossary →Tactile AI is the sensing, data, and control workflow that turns touch signals into useful robot behavior. It can support grasp confidence, slip response, contact-aware motion, safety reflexes, and evaluation analytics for Physical AI systems.
Browse tactile AI research →Physical AI is a broad term for AI systems that perceive, reason, and act through physical machines. A complete system connects sensors and models to robot policies, control, actuation, safety, and measured feedback.
Read the Physical AI explainer →Independent research. Traceable sources. Clear limitations.
Our editorial standards ↗Submit research context ↗