Andrija Milojevic

TL;DR

A lot of cool demos demonstrate a robot’s locomotion and maneuver capabilities, but we still see no highly adaptive manipulation of various objects. Human-hand-like dexterity is still out of reach for most robots. This is a problem I worked on for the better part of my carries and trying to achieve with my company thtRobotics.

To understand better see Dexter, in action—one of the most adaptive robotic gripper (at time of writing this) on the market for adaptive picking of items from cluttered spaces (which is another level of complexity, that not so many finger grippers are built to handle).

What is missing?

In the recent couple of years, we have all seen many cool demos of robots achieving various things (emphasis on the word demos). From impressive locomotion capabilities, such as jumping, flipping, swinging, and imitating various human motions, to robots trying to perform household chores or those that claim they can generalize across multiple tasks. Some say the robotics field is moving toward the age of Physical AI, promising that robots will do everything autonomously. All of this seems very exciting and “cool,” especially for someone like me who has been in the field of robotics for more than 15 years (since my student days).

But what haven’t we seen much of? What haven’t robots demonstrated? What is still very hard for robots to do? What are the limits of current robots?

It seems it is still the age-old problem in robotics—one of the holy grails: adaptive gripping and manipulation of items. The robots we have seen in recent years still struggle significantly when it comes to gripping various items, especially when they are soft and squishy. It is very hard to achieve what we do without thinking: gripping various easily damaged items.

Robots still cannot grip objects with the same level of adaptability and dexterity as human hands. This still seems out of reach for many. We see many cool demos, but no robot that can effortlessly handle, for example, a tomato, then a lemon, then an onion, banana, raspberry, mushroom, blueberry, and strawberry without stopping. There is still no (or rarely seen) robotic hand—technically called a gripper (the “end-effector” or “end tool” at the end of a robotic arm)—that can pick objects of various shapes, stiffness, sizes, and weights. The especially challenging part is picking and handling various soft and squishy things, like fresh food items.

The problem of adaptive gripping of soft items has captivated me for the better part of my career. I worked on it many years, and now with thtRobotics. We solved this problem quite some time ago, but unfortunately, not all products and innovations reach the general public. This is my small attempt to raise awareness of the important problems and limits of robots, as well as how thtRobotics is unlocking the next level of robotics—far beyond what current technology can achieve.

See couple of videos of how gripper Dexter, handles various fresh food items adaptively and effortlessly:

The problem is even more challenging when it comes to picking items from a cluttered space (e.g., a bin full of pile of tomatoes). But more on this some other time…