Open-Source Humanoid Robot Asimov 1: Now Walking While Swinging Its Arms

Day 352 of an Open-Source Humanoid Robot

A humanoid robot project called Asimov 1 is being developed entirely in public. The team has made their entire R&D process transparent, logging progress through Day 352 — and their latest milestone is this: the robot can now move its arms while walking.

That might sound like a minor feature, but in the humanoid robotics world, coordinating arm movement with walking is actually a significant technical hurdle. The team frames this step as a prerequisite for "carrying and manipulating objects while moving" — meaning it's not just a flashy demo, but groundwork for practical real-world tasks.

The project is fully open-source, with code hosted on GitHub at menloresearch/asimov-1, where anyone can view, reproduce, or contribute.

Why "Walking While Swinging Arms" Isn't as Simple as It Sounds

Humans naturally swing their arms when walking — it's an unconscious balancing mechanism. But for a bipedal robot, things are far more complex. Walking itself is a dynamic balancing problem: the robot must continuously control its center of mass, gait, and joint torques to avoid falling. Once the arms start moving, they alter the overall mass distribution and momentum, introducing additional disturbances to an already delicate balance.

To allow free arm movement during walking, the control system must incorporate upper-body motion into whole-body control (WBC), rather than treating locomotion and arm swinging as two independent subsystems. Achieving both simultaneously signals that the underlying dynamics modeling and control loop have reached a meaningful level of maturity.

Why "Manipulation While Moving" Matters

The team explicitly notes that this step is aimed toward "carrying and handling objects while in motion." Most genuinely useful real-world tasks — moving boxes, delivering items, working across a space — require a robot to use its hands while moving. Stationary robotic arm manipulation is nothing new; the real challenge lies in combining movement and manipulation. Asimov 1's current progress can be seen as establishing the kinematic foundation for future capabilities like grasping and carrying objects.

The Open-Source Value of Building in Public

A defining characteristic of Asimov 1 is its "build in public" approach — iterating day by day in full public view. This is common in the software open-source community, but relatively rare in hardware and robotics, where the barriers and costs of going open-source are much higher — spanning mechanical structures, electronics, firmware, and algorithms across multiple layers.

Opening up humanoid robot development entirely creates direct value by lowering the community's cost of learning and reproduction. For researchers and enthusiasts, a project with fully accessible code and trackable development progress is far more instructive than a closed commercial product. It also gives more people the opportunity to understand the real engineering path from zero to walking to coordinated motion.

The open-source ecosystem in robotics is fundamentally different from pure software open-source. Sharing code is enough for software, but a robotics project also involves mechanical design files (CAD), PCB schematics, actuator selection and calibration parameters, real-time firmware, and other heterogeneous assets — any missing piece makes the "open-source" label largely hollow. On top of that, the cost of reproducing hardware far exceeds that of software: even with complete documentation, sourcing parts, machining structural components, and assembling and tuning the system demands significant money and time. This is precisely why leading companies like Boston Dynamics and Figure AI keep their robots as closed black boxes. Asimov 1's choice to publish its full mechanical, electrical, and software stack — along with daily development logs — carries rare community reference value in the current white-hot race to commercialize humanoid robots.

A Measured View: There's Still a Long Road Ahead

It's worth keeping perspective: "Day 352" itself shows this is a long, incremental engineering effort, and what's being demonstrated is a milestone, not an endpoint. From "walking while swinging arms" to "stably carrying objects while moving," there's still a series of far more complex capabilities to develop — grasping, force control, perception, object recognition, and more.

But for an open-source project, maintaining a continuous and transparent record of progress is itself a remarkable commitment. It lets the outside world witness the actual growth curve of humanoid robot technology driven by community effort — rather than only seeing the polished, carefully edited clips of a product launch event.

Note: Information in this article comes from the project's progress updates shared on Reddit. For authoritative details, refer to the GitHub repository.

Background: Whole-Body Control Explained

Whole-Body Control (WBC) is one of the core challenges in modern humanoid robotics research. Traditional robot control typically decomposes motion into independent subtasks solved separately, whereas WBC treats the robot as a unified dynamical system — simultaneously computing torque commands for all joints by solving a constrained optimization problem (typically quadratic programming, QP). This approach requires maintaining a complete real-time dynamic model of the robot, including the mass matrix, Coriolis forces, gravity compensation terms, and ground contact constraints. High computational cost and sensitivity to modeling errors are the primary obstacles to deployment.

Bipedal walking is itself an "underactuated system" — the robot has more degrees of freedom than actuated joints, and the contact points between the feet and the ground are its only external support. Any additional disturbance directly affects the Zero Moment Point (ZMP) or capture point, threatening overall stability.