Boston Dynamics engineer endorses rotary actuators

Alberto Rodriguez, director of robot behavior for Atlas at Boston Dynamics, used a May 21 X thread to make a narrow but consequential point about humanoid design: actuator choice can change how quickly a robot becomes useful. Rodriguez said more teams are shifting toward rotary actuators because they are more efficient and because they reduce the gap between what works in simulation and what works on hardware. Boston Dynamics has been making that case around its new electric Atlas in recent days. In a May 18 company blog, Rodriguez and two colleagues said Atlas was built to automate physically demanding work and highlighted the speed with which the team brought up new whole-body behaviors after the robot’s January debut. The company’s latest Atlas product material also says the humanoid is designed for manufacturability, reliability and serviceability. (roboticssummit.com) The thread matters because it shifts attention from model architecture and demo polish to a hardware decision that affects sensing, controls and development time all at once. ### Why are rotary actuators getting this much attention? Rodriguez tied the argument to efficiency first. In his account, rotary actuators are becoming more attractive in humanoids because they waste less energy than alternatives and make it easier to close the loop between simulation and physical performance. That fits Boston Dynamics’ broader description of Atlas as an all-electric industrial robot built for sustained work. (bostondynamics.com) Boston Dynamics’ recent Atlas material also emphasizes a design built for serviceability and fleet use, not just research demos. That makes actuator standardization and simpler joint architectures more than a lab preference; they become part of how the machine is maintained and deployed. ### What did Rodriguez mean by “implicit proprioception”? Rodriguez’s most specific claim was that rotary actuation can let a humanoid infer unexpected load changes from the actuator itself, rather than from separate force-torque sensors. (bostondynamics.com) He described a case in which a robot expecting one fridge weight can detect that the load has effectively doubled and adapt without extra calibration hardware, according to the thread cited in the social briefing. (bostondynamics.com) Boston Dynamics’ own recent Atlas writing uses similar language about whole-body coordination and proprioception. In a separate company post on large behavior models, the firm said Atlas policies use camera inputs along with proprioception, and in the fridge-lifting write-up Rodriguez and colleagues described the robot managing heavy objects with its arms, legs and torso rather than relying only on fingertip-style manipulation. ### Why does that matter for sim-to-real transfer? (bostondynamics.com) Boston Dynamics said on May 18 that Atlas’ fridge-lifting sequence was a deliberate experiment to show advances in both hardware and behavior. The company said the robot had demonstrated gains in strength, mobility and whole-body control within weeks of its public debut in January. If actuator behavior is easier to model and measure directly, the robot needs fewer compensating sensors and less calibration work before a learned policy can survive contact with the real world. (bostondynamics.com) That is an inference from Rodriguez’s thread and Boston Dynamics’ own emphasis on proprioception and fast behavior bring-up, rather than a separately stated company claim. ### How does this fit Boston Dynamics’ current Atlas push? Boston Dynamics unveiled the product version of Atlas at CES on Jan. 5 and said 2026 deployments were already committed to Hyundai and Google DeepMind, with additional customers planned for early 2027. (bostondynamics.com) The company said the robot has 56 degrees of freedom, fully rotational joints and a 50-kilogram lift capacity. Rodriguez is also scheduled to appear at the Robotics Summit in Boston on May 27 in a keynote on the state of humanoids. (bostondynamics.com) That gives Boston Dynamics another venue, beyond the company blog and X thread, to explain why actuator architecture is becoming part of the public argument over how humanoids get from demos to deployed systems. (roboticssummit.com) (bostondynamics.com)