AI start-up raises $150M to teach robots human-like object handling
What's the story
Eka Robotics, a physical AI start-up co-founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, has raised $150 million in a Series A funding round. The investment was led by venture capital firms a16z and F-Prime. Following this round, the company is now valued at $1.45 billion post-money.
Technological advancement
Eka Robotics has surpassed human performance in some tasks
Eka Robotics has developed a manipulation architecture that includes a force-sensitive gripper and a new kind of deep-learning model.
The company's technology has already surpassed human performance in some dexterous manipulation tasks.
Raghu Raguram, general partner at a16z who co-led the funding round, said, "They are already exceeding superhuman performance for certain tasks."
He added that he expects to see superhuman performance in many dexterous manipulation tasks over the next few months.
Industry reaction
Impressive capabilities have surprised veteran robot builders
The capabilities of Eka Robotics have surprised even veteran robot builders.
Kris Hauser, a professor at the University of Illinois Urbana-Champaign, was impressed by the "wow factor" of the company's robot.
He noted that it could adapt to a wide variety of objects without retraining and displayed a "creepy humanness" in its attempts if it failed initially.
Performance metrics
Robot can pick up 1,500 objects in an hour
Eka's robot can pick up some 1,500 objects weighing up to a kilogram each in an hour.
This is a significantly better performance than humans who can do the same in about 600 picks per hour.
The company claims its technology reliably picks up objects over 95% of the time, demonstrating a high level of efficiency and accuracy.
Innovative strategy
Co-founder explains unique approach to robotic manipulation
Eka Robotics has taken a unique approach to robotic manipulation by using a vision-force-action model instead of a vision-language-action model.
This new strategy focuses on the relationship between language, force inputs, and actions.
Co-founder Agrawal explained that "vision tells you roughly where the object is, what the pose is, but the actual grabbing, the actual searching is happening through forces."