ABB Robotics has partnered with California-based bionics company Psyonic, in a project to develop more dexterous industrial robots and accelerate the adoption of physical AI in manufacturing and logistics environments.
The collaboration aims to address one of the biggest challenges facing physical AI: teaching robots to handle objects with the same dexterity as humans.
Psyonic provides its Ability Hand, originally developed as a prosthetic device, to the partnership.
Combining touch sensing, vibration feedback and articulated finger movement, the device is used by humans to generate training data for their robotic counterparts. Using the data, robots can learn complex grasping and handling tasks that have traditionally been difficult to automate.
Physical AI refers to AI systems that can perceive, reason and act in real-world environments, enabling robots to interact with objects and autonomously adapt to changing conditions rather than following pre-programmed instructions.
“Human dexterity and the instinctive understanding of how to handle different objects is one of the most difficult things to replicate in industrial-grade robotics, but it’s a fundamental need for truly autonomous and versatile robots,” Marc Segura, President of ABB Robotics, said in a release. “As we develop the next generation of physical AI, robots will learn and understand the world as we do.”
“Dexterous manipulation is ultimately a data challenge as much as a hardware challenge,” Psyonic founder and CEO Aadeel Akhtar added. “By using the same Ability Hand on people and on robots, we can capture high-fidelity real-world data on movement, contact and grip force, then use that to train robotic systems more effectively.”
The companies will integrate the technology with ABB’s collaborative robot platform, GoFa, to evaluate applications across sectors including automotive, aerospace, packaging, logistics and life sciences.
The deal reflects a broader industry push to develop physical AI systems that can operate reliably in unstructured environments.
While advances in generative AI have been driven by vast amounts of internet-scale data, robotics developers are increasingly seeking ways to collect real-world data to better train machines for physical deployment.
