Meals dealing with is an space of labor that also depends closely on people. Fruit, greens, meat, and different meals should be dealt with rapidly however gently. It’s also onerous to automate as a result of no two items of fruit, greens, or rooster nuggets look precisely the identical.
Eka’s demos recommend that the corporate could also be onto one thing huge. I discovered myself mentally evaluating their robots to GPT-1, OpenAI’s first massive language mannequin, developed 4 years earlier than ChatGPT. GPT-1 was usually incoherent however confirmed glimmers of basic linguistic intelligence.
The robots I noticed appear to have the same form of nascent bodily intelligence. After I watched a video of 1 reaching for a set of keys in gradual movement, I seen it did one thing that appeared remarkably human: It touched the information of its grippers to the desk and slid them alongside the floor earlier than making contact with the keys and securing them between its digits. Eka’s algorithms appear to know instinctively the best way to get better from a fumble. This sort of factor is tough for different robots to be taught, until the people coaching them intentionally make a variety of errors.
Not like with some other robotic I can consider, it’s virtually doable to think about what the world is like for the robotic. Its sensors appear to really feel the load of its arm, the inertia because it sweeps towards the keys and slows down. As soon as it has the keys in its grasp, it appears to sense the load of them dangling from its claw.
I don’t know if Eka’s method actually is the path to a ChatGPT-like breakthrough in robotics. Some very good specialists consider that mixing human demonstration with simulation will yield higher outcomes than simulation alone. Perhaps some mixture of the 2 approaches will finally be vital? But it surely does appear clear that robots will finally must have the form of tactile, bodily intelligence that Eka is engaged on if they’re to acquire humanlike dexterity.
Agrawal tells me that the identical basic method ought to work for finer manipulation. The fiddly dexterity required to construct an iPhone, as an illustration, may very well be achieved by constructing completely different actuators and sensors and training the duty in simulation.
After spending a couple of hours at Eka, I determine to cease by the restaurant downstairs. I watch from the counter because the workers put together meals and make espresso. A descendant of the machine upstairs could possibly do this stuff simply as nicely, if not higher. However given how a lot I take pleasure in chatting with the individuals who work there, I feel I’d pay further to maintain people round. Until, that’s, my palms get automated away too.
Tell us what you consider this text. Submit a letter to the editor at [email protected].
