Telerobotics combines human intelligence with robotic precision, repeatability and power, so people can manipulate objects in places they can’t safely reach: space, underwater, mines, hazardous sites and operating rooms.
In the IRiS Lab, we develop stability-guaranteed control, immersive haptic interfaces and physical AI that transfers human contact-rich manipulation skills to robots. Our goal is stable, transparent, high-performance haptic and telerobotic systems, from industrial operations to medical procedures.
We capture how people feel contact, turn those signals into representations a robot can learn from, and transfer the resulting skills back to robots, with haptic feedback throughout.
How do we record a human demonstration, including what the person feels?
Handheld devices have made robot data collection cheaper, but they don’t record touch. We are building handheld devices that also capture haptic signals, and studying how what a person feels relates to what the sensors observe, so that robots can learn human skills from these recordings.
General-purpose master devices such as Phantom, Omega or Virtuose limit how intuitive teleoperation can be: the operator loses situational awareness, and their hand-eye coordination doesn’t match the remote robot. We build task- and robot-specific haptic interfaces instead, such as bimanual haptic displays made from collaborative robot arms and modular 3-DoF torque feedback devices that attach to the tip of commercial haptic devices.
Model-Free Energy-Based Friction Compensation for Industrial Collaborative Robots as Haptic Displays
IEEE/ASME T-MECH, 2025
CoaxHaptics-3RRR: A Novel Mechanically Overdetermined Haptic Interaction Device Based on a Spherical Parallel Mechanism
IEEE Telepresence, 2024
Optimizing Setup Configuration of a Collaborative Robot Arm-Based Bimanual Haptic Display for Enhanced Performance
IEEE RA-L, 2024Rendering stiff virtual environments is hard for haptic devices. Time discretization, delay, encoder quantization and the zero-order hold of the force command all inject energy, and the contact becomes unstable unless friction or the user’s grip dissipates it.
We have proposed several stability-guaranteed approaches, including the time-domain passivity approach (TDPA), input-to-state stability (ISS), successive force augmentation (SFA), successive stiffness increment (SSI) and unidirectional virtual inertia (UVI). We keep extending the range of impedances a haptic system can render transparently.
Relaxing Conservatism for Enhanced Impedance Range and Transparency in Haptic Interaction
IEEE Transactions on Haptics, 2024
Enhancing Contact Stability in Admittance-Type Haptic Interaction Using Bidirectional Time-Domain Passivity Control
IEEE RO-MAN, 2023
Error-Domain Conservativity Control to Transparently Increase the Stability Range of Time-Discretized Controllers
IEEE ICRA, 2023
Chattering-Free Time Domain Passivity Approach
IEEE Transactions on Haptics, 2022
Virtual Inertia as an Energy Dissipation Element for Haptic Interfaces
IEEE RA-L, 2022
Successive Stiffness Increment and Time Domain Passivity Approach for Stable and High Bandwidth Control of Series Elastic Actuator
IEEE ICRA, 2020
Time-domain passivity control of haptic interfaces
IEEE Transactions on Robotics and Automation, 2002How do we turn raw touch signals into something a robot can learn from?
Robot learning still makes little use of haptic data, largely because there is no common way to represent it. We are developing learned haptic representations, informed by physics and by how humans perceive touch, to improve robot learning.
Direct teleoperation is physically and mentally demanding and needs trained operators. We add autonomy to support them. Virtual fixtures guide the tool, follow a path or keep it out of dangerous regions; we generate them interactively in unstructured environments and adapt how they are rendered to time delay and task type.
At the other end of the scale, the operator guides an autonomous planner through a haptic device, which simplifies motion planning in cluttered environments.
Haptic Field and Force Feedback Generation for Wheeled Vehicle Teleoperation on 2.5D Environments
Lecture Notes in Networks and Systems, 2024
Method for generating real-time interactive virtual fixture for shared teleoperation in unknown environments
The International Journal of Robotics Research, 2022
Simultaneous Use of Autonomy Guidance Haptic Feedback and Obstacle Avoiding Force Feedback for Mobile Robot Teleoperation
ICCAS, 2022How does a robot learn a dexterous, contact-rich skill from these demonstrations?
Haptic data is underused when robots learn contact-rich tasks. We extract dexterous skills from human demonstrations using high-fidelity tactile information.
Learning from demonstration trains robot skills with far less data than reinforcement learning or large vision-language-action models need. Our immersive haptic teleoperation system lets the operator demonstrate a task bare-handed, with vision, force and tactile feedback, so the robot learns from the person’s actual motions and intentions.
Is a Simulation better than Teleoperation for Acquiring Human Manipulation Skill Data?
IEEE/RSJ IROS, 2024
Learning Robotic Rotational Manipulation Skill from Bilateral Teleoperation
IEEE UR, 2022
Mini-Batched Online Incremental Learning Through Supervisory Teleoperation with Kinesthetic Coupling
IEEE ICRA, 2020
