An Interview with Ken Goldberg: Exploring Robotics at UC Berkeley

Ken Goldberg, a professor and the William S. Floyd Jr. Distinguished Chair in Engineering at UC Berkeley, robotics entrepreneur, and IEEE fellow, believes that the transformative potential of generative AI in robotics is only just beginning to be realized in 2023. He envisions that large language models like ChatGPT can help robots and humans communicate naturally and effectively. Additionally, advancements in Vision-Language-Action models have shown promise in enhancing robot perception and control, although the need for vast data sets and questions about generalization remain.

Goldberg also sees the emerging field of “Multi-Modal models” as a significant area of interest, particularly in the combination of different input modes and actions in response to the same input state. He highlights the potential of Diffusion Policies in preserving multi-modal actions, reinforcing the strides being made in robotics.

Regarding the humanoid form factor in robotics, Goldberg expresses his evolving perspective, acknowledging recent technological advancements in humanoids and quadrupeds as impressive. He recognizes the potential advantages of legged robots over wheeled counterparts in traversing complex environments and foresees their increasing utility in homes and factories.

Amidst the prominence of robotics in manufacturing and warehouses, Goldberg points to robot-assisted surgery and self-driving taxis as other promising areas, albeit with reservations about their cost-effectiveness. He raises the concept of “Augmented Dexterity” in surgical robotics as a subject of interest and notes the potential for further advancements in these fields.

When it comes to the timeline for the development of true general-purpose robots, Goldberg tempers expectations, emphasizing that achieving true AGI and general-purpose robots is unlikely to occur in the near future. He dismisses concerns about robots overtaking human jobs and foresees the gradual integration of home robots beyond vacuum cleaners, envisioning their role in decluttering tasks.

In his opinion, robot motion planning is an underreported aspect of robotics, specifically citing the persistent challenge of robot singularities. Goldberg details the detrimental impact of singularities on robot operation and introduces Jacobi Robotics, a startup he co-founded, as a solution that implements algorithms to avoid these disruptions, potentially increasing reliability and productivity for all robots.

Overall, Goldberg’s insights underscore the ongoing evolution and advancement of robotics, with generative AI, multi-modal models, and the ongoing pursuit of efficient motion planning occupying a central place in shaping the future of robotics. As the field continues to progress, Goldberg’s expertise provides valuable perspectives on the current landscape and the untapped potential of robotics.

Share:

Related News