Axol

Axol is a dual-arm robot purpose-built for builders working on physical AI. As a complete, ready-to-deploy dual-arm robot, it empowers researchers and engineers to automate physical tasks without weeks of setup. Designed by Almond and backed by Y Combinator, each unit is assembled in the Dogpatch neighborhood of San Francisco, California, emphasizing hands-on craftsmanship. The robot's core value proposition is eliminating compromises: its seven-degree-of-freedom arms offer 860mm reach and 6.5kg peak payload, while an open-source SDK and VR teleoperation pipeline provide a seamless path from experimentation to production. This dual-arm robot addresses the gap between research platforms and industrial robots, offering a capable, open, and affordable solution starting at $7,999.

The concrete problem Axol solves is the limited workspace and singularities common in comparable dual-arm platforms. Many robots force users to reposition frequently or accept constrained trajectories, slowing down development and deployment. Axol's extra-long reach of 860mm per arm significantly outperforms alternatives like YAM (750mm) and WidowX (650mm), giving access to a wider area without moving the base. Additionally, full 180° pitch and yaw at the wrist dramatically reduces shoulder singularities, resulting in smoother, more reliable motions. This matters because physical AI tasks often involve complex, contact-rich manipulation over large volumes; Axol's design ensures fewer interruptions and higher throughput in production environments.

The first major feature group is the dual 7-DOF arms themselves. Each arm is independent with 860mm reach and 6.5kg peak payload, operated via USB-C CAN control at 500Hz. This high-bandwidth, low-latency interface enables precise joint-level control, critical for training and deploying learned policies. The two arms can work together for bimanual tasks or double the cycle time of a single-arm robot by operating on separate workpieces. Protected wiring with fully internally routed cables ensures cables are not damaged during motion, while modular wrists accept custom end-effectors or grippers. These design choices make the robot robust for repeated, long-duration use in real-world settings.

The second major feature group is VR teleoperation via WebXR. Axol supports teleoperation from any compatible headset, streaming hand and elbow poses over WebSocket. The system includes built-in data collection and recording modes, allowing operators to demonstrate tasks for imitation learning. This is useful because it provides an intuitive way to generate high-quality training data without programming complex motion sequences. Users can control the robot in real time, with natural arm movements translated to bimanual actions. The VR teleop pipeline is open source and integrated with the larger SDK, making it easy to extend or customize for specific data collection needs.