For the next several weeks, TechCrunch’s robotics newsletter, Actuator, will be running Q&As with some of the top minds in robotics. Subscribe here for future updates.
Part 1: CMU’s Matthew Johnson-Roberson
Part 2: Max Bajracharya and Russ Tedrake of the Toyota Research Institute
Part 3: Dhruv Batra of Meta
Part 4: Aaron Saunders of Boston Dynamics
Ken Goldberg is a professor and William S. Floyd Jr. Distinguished Chair in Engineering at UC Berkeley, co-founder and chief scientist at robotic parcel sorting startup Ambidextrous and IEEE Fellow.
What role(s) will genetic artificial intelligence play in the future of robotics?
Although the outcry started a little early, 2023 will be remembered as the year genetic AI transformed Robotics. Large language models like ChatGPT can allow bots and humans to communicate in natural language. Words have evolved over time to represent useful concepts from “chair” to “chocolate” to “gift.” Roboticists also discovered that the great Vision-Language-Action Models can be trained to facilitate robot perception and control the robot’s arm and leg movements. Training requires massive amounts of data, so labs around the world are now collaborating to share data. The results are flooding in, and although there are still open questions about generalizability, the impact will be profound.
Another fascinating topic is “Multi-Modal Models” with two meanings of multi-modal:
- Multi-Modal in the combination of different modes of input, e.g. Vision and Language. This is now being expanded to include sense of touch and depth and robot actions.
- Multi-Modal in terms of allowing different actions in response to the same input state. This is surprisingly common in robotics. for example there are many ways to grab an object. Typical depth models will “average out” these capture actions which can produce very poor capture. A very exciting way to preserve multimodal actions is Diffusion Policies, developed by Shuran Song, now at Stanford.
What are your thoughts on the humanoid form factor?
I’ve always been skeptical of humanoids and legged robots as they can be overly impressive and ineffective, but I’m reconsidering after seeing the latest humanoids and quadrupeds from Boston Dynamics, Agility and Unitree. Tesla has the engineering skills to develop low-cost engines and powertrains at scale. Legged robots have many advantages over wheels in homes and factories for traversing steps, debris and carpets. Bi-manual robots (with two hands) are essential for many tasks, but I still believe that simple grippers will continue to be more reliable and cost-effective than five-fingered robot hands.
After manufacturing and warehouses, what is the next big category for robotics?
After the recent union wage settlements, I think we will see a lot more robots in manufacturing and warehouses than we have today. Recent progress in self-driving taxis has been impressive, especially in San Francisco where driving conditions are more complicated than Phoenix. But I’m not convinced they can be cost-effective. For robot-assisted surgery, researchers are exploring ‘Augmented Dexterity’ – where robots can improve surgical skills by performing low-level secondary tasks such as suturing.
How far are true general purpose robots?
I don’t expect to see true AGIs and general purpose robots in the near future. Not a single roboticist I know is worried about robots stealing jobs or becoming our masters.
Will home robots take off (beyond the gaps) in the next decade?
I predict that within the next decade we’ll have affordable household robots that can shuffle — pick up things like clothes, toys, and trash off the floor and place them in appropriate bins. Like today’s vacuum cleaners, these robots will occasionally make mistakes, but the benefits for parents and the elderly will outweigh the risks.
What major robotics story/trend isn’t getting enough coverage?
Robot motion design. This is one of the oldest topics in robotics – how to control motor joints to move the robot tool and avoid obstacles. Many believe that this problem has been solved, but it is not.
Robot “peculiarities” are a fundamental problem for all robot arms. they are very different from Kurzweil’s hypothetical moment in time when AI surpasses humans. Robot singularities are points in space where a robot stops unexpectedly and must be manually reset by a human operator. The singularities result from the mathematics required to convert the desired rectilinear motion of the handle into the corresponding motions for each of the six robot joint motors. At some point in time, this conversion becomes unstable (similar to a divide-by-zero error) and the robot needs to reset.
For repetitive robot motions, singularities can be avoided by tedious manual fine-tuning of repetitive robot motions so that they never encounter singularities. Once such moves are identified, they are repeated over and over again. But for the growing generation of applications where the robot’s movements are not repetitive, including palletizing, bin picking, order fulfillment and package sorting, singularities are common. It’s a well-known and fundamental problem as they interrupt the robot’s operations at unpredictable times (often several times an hour). I co-founded a new startup, Jacobi Robotics, this one implements efficient algorithms that are *guaranteed* to avoid singularities. This can significantly increase reliability and productivity for all robots.
