Boston Dynamics’ Atlas Lead on the Humanoid Form as a ‘Software Problem’

In a wide-ranging interview with @TheHumanoidHub , Boston Dynamics’ Atlas Product Management Lead, Mario Bollini, provided key insights into the company’s strategy, framing the humanoid form not as an end in itself, but as a way to "shunt complexity" from hardware to software. Bollini touched on the robot’s superhuman capabilities, the pivotal shift to electric actuation, and the lessons learned from commercializing its quadruped, Spot.

The Humanoid as a General Platform

Addressing the ongoing industry debate about the utility of the humanoid form factor, Bollini argued that it provides crucial generality, especially in the early stages of product development. "For us, having a humanoid form factor allows us to turn...customer to customer requirements into a software problem," he explained.

This approach gives Boston Dynamics the "maximum aperture" to explore commercial opportunities and find the shortest path to value for customers. While more specialized robots with wheels or different arm configurations might eventually prove superior for specific, mature tasks, the bipedal, two-armed humanoid is the ideal platform for discovering those applications in the first place.

Superhuman by Design

A key theme of the discussion was that the new electric Atlas is not strictly bound by human limitations. Bollini highlighted features like the robot's ability to rotate its torso and legs with a range of motion far exceeding a person's. This was not an accident, but a "deliberate engineering decision."

These "superhuman" capabilities create tangible product advantages. For example, a robot that can spin its torso 180 degrees instead of taking several steps can reduce cycle times in a manufacturing setting, directly adding value for a customer. "It's having a huge impact on our product pipeline by allowing Atlas to move faster than a person," Bollini noted.

An Evolution in Brains and Brawn

The development of the new Atlas was enabled by what Bollini described as a "point of convergence" in technology. Advances in battery density and the company's decision to design its own high-performance electric motors and actuators in-house finally made it possible to move away from the powerful but messy and potentially dangerous hydraulic systems of the previous generation.

The control software has seen an equally dramatic paradigm shift. While the old Atlas relied on classical control approaches like Model Predictive Control (MPC), the new platform has fully embraced modern machine learning. Bollini said that reinforcement learning (RL) and behavior cloning have "dramatically accelerated the timeline" for developing both locomotion and manipulation skills.

This philosophy of pragmatic, application-driven design extends all the way to the robot's end-effectors. As we recently covered, the company opted for a three-fingered hand that balances dexterity and reliability, another decision that prioritizes real-world utility over mimicking human anatomy.

Watch the Full Interview

Bollini's discussion offers a clear window into the strategic thinking that is shaping one of the industry's most-watched humanoid projects. For a deeper dive into these topics, you can watch or listen to the full interview below.