Could a mass‑produced humanoid robot outwalk most humans without ever stopping? The AgiBot A2 has just provided a resounding answer, completing a Guinness World Record‑setting 66‑mile trek from Suzhou’s Jinji Lake to Shanghai’s Bund in a single, uninterrupted march. Over nearly three days, the 1.75‑meter‑tall, 55‑kilogram biped navigated highways, bridges, tiled pavements, slopes, and low‑light urban corridors without a single shutdown thanks to an autonomous hot‑swappable battery system that kept its actuators and processors powered continuously.
The A2’s energy management architecture is central to this achievement. Like the industrial robot, UBTech’s Walker S2, the system allows for the removal of spent batteries while continuing operations. This eliminates the downtime that would be associated with recharging the system’s batteries between endurance demonstrations. Without stopping to recharge, the A2 could complete endurance demonstration events without interruption and continue to operate stably throughout the entire endurance. In addition, the system was able to maintain a consistent voltage level throughout the battery swapping process; this prevented voltage drops that could have disrupted the A2’s control and destabilized its gait. This is an important aspect of the A2’s design since many bipedal locomotion systems depend on careful modulation of torque at multiple joints, thus requiring very close control of motion.
A diverse route requires more than GPS coordinates, therefore the A2 has two GPS modules and has integrated three different technologies into one stack. By using LIDAR and infrared, we can determine where the robot is located (e.g., under a bridge) if it’s receiving poor signals from satellites (e.g., no satellite data). Using LIDAR mapping creates a high-resolution map of the environment, while infrared depth sensors enable the robot to see at night. When combining all of these sensors, along with a real-time algorithm for determining the safest path, the robot can accurately avoid trenches and steep heights and move away from vehicles and people while it performs its tasks with no human intervention.
Maintaining stability over 106 kilometers demanded advanced locomotion control. The A2’s balance algorithms, refined through hundreds of hours of continuous walking trials, draw on principles similar to those used in research platforms such as Cassie, which employs machine learning to optimize gaits for speed and endurance. While Cassie’s record‑setting sprint showcased dynamic balance under high acceleration, the A2’s feat emphasized sustained stability under cumulative mechanical stress. Each step required micro‑adjustments to joint angles and force distribution, compensating for surface irregularities and gradual wear on its rubber foot soles.
The hardware endurance demonstrated in this walk is notable for a commercial unit. AgiBot confirmed that the robot was identical to those delivered to clients no reinforced joints, upgraded batteries, or special firmware. After three days of continuous operation, the only visible wear was on the outer rubber layer of its soles. This resilience underscores progress in actuator durability, thermal management, and mechanical tolerances, all critical for deploying humanoids in real‑world service roles.
Throughout the journey, the A2 complied with traffic regulations, an operational constraint that tested its decision‑making algorithms. The algorithms controlling autonomous vehicle behavior must use the sensory input to follow the rules of movement established by law. The rules include stopping at crosswalks, yielding to other vehicles, and so forth, which defines how the autonomous vehicle will behave on the roadway. Combining these types of rules with real-time (reactive) avoidance algorithms demonstrates another growing trend within the field of humanoid robot design, that being the incorporation of social and regulatory compliance within the navigation stack so that the robot can operate safely within a human-occupied area.
Upon arrival, the robot retained enough processing bandwidth and battery reserve to interact with reporters, calling the trek a “memorable experience” in its “machine life” and joking about needing “a new pair of shoes.” In addition to being funny, this example illustrates how the A2 uses AI to perform conversation management through its speech synthesis capabilities in different languages, as well as its ability to analyse and understand people’s faces and recall past conversations with users. These features along with their experience working with others give the A2 an excellent foundation to build applications in multiple industries, including logistics, customer support and automated delivery methods such as drones and autonomous vehicles.
Industry observers see this as a milestone in environmental adaptability. As Liu Dingding noted, such achievements signal improvements in “robot reliability, endurance, motion control, and environmental adaptability,” and suggest that humanoids may soon match or surpass humans in certain physical tasks. The A2’s record walk is not just a publicity stunt it is a data‑rich validation of integrated energy management, sensor fusion navigation, and stability control in mass‑produced humanoid robots, marking a step toward autonomous machines capable of operating alongside humans without constant supervision.
