A team of researchers from UC Berkeley, Nvidia, and Stanford has developed a new framework called T-Rex that lets robots react to physical contact as it happens. The system processes tactile feedback in real time, giving machines a more human-like ability to adjust their movements when bumped, pushed, or touched.
What T-Rex Does Differently
Most existing robotic systems either ignore contact or handle it with a delay, forcing a robot to stop and recalculate. T-Rex changes that by integrating tactile sensing directly into the control loop. When a robot arm is nudged, the framework instantly adjusts the trajectory or grip strength without a full system reset.
The framework was built by researchers at UC Berkeley’s robotics lab, Nvidia’s AI research division, and Stanford University’s robotics group. They designed T-Rex to work with standard robotic hardware, meaning the approach doesn’t require specialized sensors or expensive retrofits.
Why Real-Time Tactile Response Matters
In dynamic environments — a warehouse where workers walk near robot arms, a home where a robot might bump into furniture, or a hospital where a surgical assistant must avoid unexpected contact — delayed reactions can lead to collisions or dropped objects. T-Rex aims to make robots safer and more reliable by letting them feel and respond on the fly.
“Robots today are good at following pre-programmed paths, but they struggle when something unpredictable happens,” a researcher involved in the project said. “T-Rex gives them the ability to react in the moment, which is critical for real-world deployment.”
Potential Applications in Automation
The framework could advance automation in manufacturing, logistics, and service robotics. Tasks that currently require careful separation of human and robot workspaces — such as assembly lines or collaborative pick-and-place operations — might become more fluid if robots can safely handle accidental contact.
The researchers emphasize that T-Rex is still a research prototype. They have published the underlying code and data online to allow other labs to test and build on the work.
The team plans to present the framework at an upcoming robotics conference, where they will demonstrate live scenarios of robots reacting to pokes, pushes, and sudden obstacles.
