See related:
Demise of Dumb: Why Make Industrial Robots Smart? A new generation of robots is on the rise, and their calling card reads, smart!he demise of dumb

“In my view,” notes Wang, the study’s lead author, “another big problem is that the data come from so many different domains, modalities, and robot hardware. Our work shows how you’d be able to train a robot with all of them put together.”

“The main advantage of this technique is its ability to integrate data from different sources into a unified system. This approach is similar to how large language models are trained, showing proficiency across many tasks due to their extensive and varied training data. HPT enables robots to learn from a wide range of experiences and environments.”

Robotis plans to build physical AI capabilities into its cobot arm and gripper, called the OpenMANIPULATOR-Y (OM-Y), Currently, OM-Y performs tasks with a gripper-type end-effector, but with human-level manipulation capabilities, it is expected to operate more efficiently in a wider variety of industrial environments.

“Developing physical AI requires cutting-edge precision components, including tactile sensors that mimic the touch sensitivity of human hands and ultra-small, ultra-precise actuators equipped with high back-drivability and torque density.”

Robotis also plans to utilize its actuator technology to collaborate with MIT researchers to develop technologies and components related to physical AI.

Physical AI is all about systems on a robot that learn about an environment directly from sensor data, like sensors and actuators, including cameras, microphones, temperature gauges, inertial sensors, RADAR and LiDAR, physical agents that enable robots to solve problems that involve direct interaction between them and the physical world.

Our elusive sixth sense…in a machine
Maybe the toughest challenge for robot training and HPT is proprioception, better known as our sixth sense. “Sight, hearing, smell, taste, touch: We’re all familiar with the five senses that allow us to experience our surroundings,” says Dr. Niccolò Zampieri, head of the Development and Function of Neural Circuits Lab at the Max Delbrück Center in Berlin.

“Equally important but much less well known is the sixth sense: Its job is to collect information from the muscles and joints about our movements, our posture and our position in space, and then pass that on to our central nervous system. This sense, known as proprioception, is what allows the central nervous system to send the right signals through motor neurons to muscles so that we can perform a specific movement.

“People [as well as robots] without proprioception can’t actually perform coordinated movements.” Like how to put a cup to your lips in the dark.

Lirui Wang and his crew at MIT are well aware of their program’s need to address some aspect of proprioception for their HPT training techniques. Until then, bipedal humanoid robots will look gawky and stiff while walking or while attempting any coordinated movement. Same goes for cobot arms while tending a CNC machine.

There’s certainly a ways to go for physical AI and robotics to totally converge, but the journey is afoot.