Tactile Sensing Emerges as the Next Robotics Frontier

For decades, robots have been able to see, hear, and move with increasing sophistication. But one crucial human sense has remained frustratingly out of reach for machines: touch. Now, a wave of new research from academic labs and startups is finally giving robots the ability to feel, and it could unlock one of the last great frontiers in automation — dexterous manipulation.

Dexterous manipulation — the ability to handle objects with the fine-grained control that humans take for granted — remains one of the biggest barriers preventing robots from successfully tackling a wide range of everyday tasks. From plugging in a USB cable to turning a key in a lock, tasks that seem trivial to humans remain maddeningly difficult for even the most advanced robotic systems. The missing ingredient, a growing consensus of researchers now believes, is tactile feedback.

The Limits of Vision-Only Robotics

Over the last few years, vision-language-action (VLA) models have significantly improved the ability of robots to carry out complex tasks involving objects and environments they have never encountered before. Pretrained on huge amounts of images, video, and text, and then fine-tuned on a smaller number of teleoperated robot demonstrations, these VLAs can guide robots through a growing range of everyday jobs — folding laundry, tidying living rooms, and even operating kitchen gadgets — using just a video feed and natural language instructions.

But robots still struggle with tasks that require fine-grained hand control, such as handling deformable materials or manipulating small objects. That is partly because VLAs ignore one of the primary sources of information humans rely on in these situations: the sense of touch.

“Most dexterous manipulation can be done by humans with their eyes closed,” says Trevor Darrell, professor of computer science at the University of California, Berkeley. “Understanding force, slip, and precise grasping is not something that can be done well with traditional vision sensors.”

The T-Rex Dataset: A Sense of Touch for Robots

Making effective use of tactile sensors is difficult. Tactile sensor data has very different characteristics compared to the image data that VLAs are normally trained on, and tactile datasets lag far behind the internet scale of many vision and language datasets. To get around this, Darrell’s team at UC Berkeley devised a way to first pretrain a model on existing datasets before giving it a sense of touch by training a specialist submodel on 100 hours of specially collected, high-quality tactile data. This data included demonstrations of common actions like wiping, grasping, twisting, or pouring using more than 200 different household objects.

Putting the tactile data to use was not straightforward. The goal was for a robot to be able to use the tactile signal to correct its grip in real time as it manipulated objects. But this requires reaction times faster than most vision-language models operate at. This mismatch is a significant challenge, so the team used separate submodels — known as “experts” — to handle high-level actions and low-level tactile control in a way that is quick enough for the tactile feedback to be useful.

The action expert produces motion plans, while the tactile expert, which operates four times faster, uses tactile feedback to adjust the motion plan in real time based on what the robot is feeling. The model was then fine-tuned on about 100 teleoperated demonstrations of relatively complex manipulation tasks, such as screwing in a light bulb, applying toothpaste to a toothbrush, or transferring an egg between trays. The results were striking: it averaged a success rate of 65 percent across 12 tasks — nearly double the best VLA model.

The Hardware-Agnostic Approach

One limitation of the Berkeley approach is that the data comes from a single instance of robotic hardware. Robot hands range from fully articulated five-finger designs to simple pincer grippers, and tactile sensors can rely on fundamentally different physics, from measuring changes in resistance to recording images of a soft gel pad deforming. This makes most tactile AI research sensor-specific, says Chengbo Yuan, a master’s student at Tsinghua University in Beijing, and makes it hard to share data and transfer learnings between research groups.

Yuan recently set out to tackle this problem by aggregating more than 3,000 hours of tactile robotic data from publicly available datasets, covering 21 sensor types and a variety of robot embodiments. The team designed a hardware-agnostic model that can train on this diverse data by converting each sensor’s output into a shared format and mapping it onto labeled positions on a template of a human hand. This model was much more successful than a baseline model, even on hardware it had never encountered before. Yuan attributes this to the model acquiring “some kind of common sense of tactile knowledge” by training on such diverse setups.

Scaling Up: The Race for More Data

Despite the promising results, Yuan thinks more tactile data is needed. His group is now leading an 80-institution collaboration to collate a larger set of teleoperated demonstrations using a standardized approach to tactile data collection and processing.

In the meantime, Fudan University in Shanghai and its spin-out NeoteAI have already produced a tactile dataset an order of magnitude larger than previous efforts. Using a proprietary sensor attached to a variety of robotic arms and a handheld gripper operated by humans, they have collected more than 30,000 hours of demonstrations with synchronized visual and tactile data.

The researchers used this data to train a model that does not just react to touch but also proactively predicts what the robot should be feeling to help guide and assess actions, significantly improving performance. Shunlin Lu, a postdoc researcher at Fudan University and CTO of NeoteAI, says the results are clear evidence that access to large-scale and diverse tactile data leads to significant performance gains.

Inferring Touch From Vision

Another approach to scaling tactile data could be to piggyback on the vast quantities of visual robotics data already collected. Researchers at the University of Southern California in Los Angeles recently released a model that learned to infer tactile information from visual data. They trained it on more than 2,700 demonstrations of everyday manipulation using a handheld gripper that records both tactile data and images from a camera on the device.

The model learned associations between images of the gripper coming into contact with objects and the amount of pressure felt by the tactile sensors at that moment. This gives even robots without tactile sensors a rudimentary sense of touch that the researchers showed to be particularly useful for contact-rich manipulation tasks. Their broader ambition is to use this generator to add tactile data to existing vision datasets — potentially transforming the entire landscape of robot training data overnight.

Predicting Surprise: The IROS 2026 Approach

How much tactile data will be required for breakthroughs in dexterous tasks remains unclear. So far, tactile training’s main contribution has been to make robots more efficient learners at tasks already within reach, like picking and placing objects. New algorithms may be required to tackle problems truly impossible without touch.

Long Cheng of the Chinese Academy of Sciences in Beijing thinks raw data alone is no panacea. “Data is good,” he says. “But how to use them correctly is another issue.” The problem is that vision provides a continuous, high-bandwidth stream of pixels, while tactile signals are sparse and intermittent, so models learn to ignore them. His solution, being presented at IROS 2026 later this month, is a model that predicts what a robot will feel from vision alone and then compares it against real tactile input. A large gap between the two means the sensor is detecting something the robot would otherwise miss, so these surprising signals are amplified while predictable ones are dampened. Across five contact-rich tasks, the approach averaged 62.8 percent success against 28.2 percent for the same model without touch.

The Road Ahead for Tactile Intelligence

Lu is confident that data scaling could have similar benefits to those seen in areas like language and vision. He guesses closer to 100,000 hours of data, collected in varied, real-world settings rather than in the lab, could unlock new capabilities. The field now has some early signs that larger tactile datasets and smarter ways to use them can give robots a significant boost on some of the most challenging tasks.

• T-Rex dataset from UC Berkeley: 100 hours of tactile data across 200+ household objects
• Tsinghua University: 3,000+ hours aggregated from 21 sensor types, hardware-agnostic model
• Fudan University and NeoteAI: 30,000+ hours with synchronized visual and tactile data
• USC Los Angeles: Inferring touch from vision across 2,700+ demonstrations
• Chinese Academy of Sciences: Surprise-based tactile signal amplification at IROS 2026

The implications extend far beyond the laboratory. Robots that can feel could transform industries from manufacturing to healthcare, enabling automated systems to handle delicate tasks like food preparation, medical procedures, and precision assembly with the same ease that humans do. The race is on, and the sense of touch may well be the key that unlocks the next era of physical AI.

As Lu puts it: “I think tactile intelligence is actually the next step for physical AI.”

---

Edited by Palawan @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous