Boston Dynamics names Alexa and Nova leader Rohit Prasad CEO

Boston Dynamics has appointed Rohit Prasad as CEO, effective immediately, the company said. He succeeds Robert Playter and is expected to join the board, subject to customary approval.

Prasad takes charge as Hyundai Motor Group prepares to move Boston Dynamics’ Atlas humanoid from trials into automotive manufacturing. His experience building large-scale AI products addresses the software required to connect capable machines to factory workflows, operate fleets and adapt robots to new tasks.

From Alexa to Atlas

Prasad spent 12 years at Amazon, most recently as senior vice president and head scientist for Alexa and artificial general intelligence. He helped scale Alexa across hundreds of millions of devices and led development of the Amazon Nova family of foundation models, which are trained on broad datasets and adapted to multiple enterprise tasks.

Before joining Amazon, Prasad spent about 14 years at Raytheon BBN Technologies, working on machine learning research and applied systems for government and commercial customers. Alexa and Nova gave him experience turning probabilistic AI into products with developer interfaces, operational controls and large deployment footprints. Robotics introduces additional requirements around physical safety, hardware reliability and real-time control.

Hyundai’s factory math

Boston Dynamics sells Spot, a four-legged robot used for inspection and data collection, and Stretch, a case-handling robot for warehouses. Atlas, its electric humanoid, is the platform behind Hyundai’s factory plans.

Hyundai acquired majority ownership of Boston Dynamics in 2021 and later bought SoftBank Group’s remaining stake, making the robotics company a wholly owned subsidiary. Hyundai has said it plans to deploy Atlas at its Georgia plant beginning in 2028 and develop capacity to produce as many as 30,000 humanoids annually.

Producing robots at that scale requires software for perception, task planning, fleet management, security, updates and factory-system integration. Prasad’s background also fits a potential recurring-revenue business built around managed software, model access and developer tools.

Physical AI needs a software stack

Boston Dynamics describes its direction as “Physical AI,” meaning software that converts sensor data and instructions into actions performed by a machine. One common approach is a vision-language-action model, which processes images and written or spoken commands, then generates robot actions. A control policy performs a similar role by mapping the robot’s current state to its next movement.

Four layers will reveal how that strategy changes Boston Dynamics’ products:

• General robot models. Models that support multiple tasks could reduce the engineering required to program each Atlas or Spot deployment.
• Developer interfaces. Stable APIs, simulators, task-authoring tools and model access would help customers connect robots to existing applications and operational systems.
• Fleet operations. Production deployments need observability, access controls, remote recovery, safety reporting and over-the-air updates.
• Factory validation. Hyundai’s Georgia plant and its Robotics Metaplant Application Center give Boston Dynamics controlled environments for training and testing Atlas on automotive work.

Humanoid rivals chase the same stack

Atlas is competing for industrial deployments with humanoids from Figure, Agility Robotics, Tesla, Apptronik, Unitree and 1X. Their technical pitches increasingly emphasize end-to-end neural policies and vision-language-action models. Competition now centers on task success rates, adaptation speed, uptime, safety, deployment cost and integration effort.

Bloomberg described the hire as part of Boston Dynamics’ push toward commercialization. Prasad’s appointment suggests that model quality, software operations and developer access will carry more weight as humanoids move into working factories.

The roadmap will surface in the tools

Boston Dynamics announced no accompanying changes to Spot or Stretch APIs, SDK licensing, support terms, Atlas access or product release dates. Prasad’s strategy will become concrete when the company publishes model interfaces, simulation and data tools, deployment controls or access terms for Atlas. The developer-platform roadmap remains unpublished.