Inside Atlas: How Humanoid Robot Design Shapes Vision, Sensors, and Human-Safe Interaction

Boston Dynamics’ webinar, Form & Function of Enterprise Humanoid Design, held March 18, touched on a range of topics, including insights on the design considerations of the robot’s vision system and sensor placement. 

According to Aaron Abroff, head of industrial design, the cameras are primarily mounted in the robot’s head to maximize its field of view and reduce occlusion by the root’s own body. The elevated camera position gives the robot a better chance to see its environment comprehensively without parts of its structure blocking the view.

James Cuseo, technical director, compute and sensing hardware, and a newer member of the design team with a background in consumer electronics, described the robot’s head as “a computer on a neck” that houses the camera array and computing hardware. The head design accommodates robust performance needs, including requirements or waterproofing and protecting sensors against impacts and environmental exposure. He noted that achieving this robustness while maintaining sensor functionality is a challenging task that will continue to be refined through future iterations.

The team highlighted that the robot’s neck includes a pitch degree of freedom, allowing the head—and therefore the cameras— to tilt. This movement is used not only for perception tasks such as viewing the robot’s own feet and nearby ground but also to look upward when reaching for objects on shelves. “I think we wanted to have some way to acknowledge when somebody waked into the space,” Abroff said. “Maybe give a command…just a little ten-degree nod.”

The design process also considered the interaction of sensors with safety features and human operation. For instance, the robot’s rear handles for manual manipulation partially obstruct the cameras’ field of view, creating occluded areas the software will have to manage. “We initially tried to design something with no compromises,” Cuseo said, “but it’s really important that the operators who might be handling a robot in the unpowered state have something safe to grab on.” These tradeoffs between safety and vision coverage were acknowledged as a necessary part of the design.

Abroff described the use of “invisible geometry” in design—modeling the projected fields of view of the cameras as three-dimensional volumes to ensure optimal sensor placement relative to mechanical constraints and workspace requirements. “We’re projecting a kind of rectangle out from the cameras that gets bigger as it gets farther away from the robot. That is so the robot can see its environment, and so that it can see its grippers,” Abroff said.

Overall, the placement and integration of cameras on the new Atlas robot balance maximizing environmental awareness with practical considerations of robot structure, operator safety, and sensor protection.