Galbot introduced PASSAGE, a research framework that helps humanoid robots navigate crowded environments by adapting their movements to obstacles.
The system allows robots to get through tight spaces in a variety of ways, including climbing over objects, ducking under barriers and turning sideways to fit through tight gaps.
The framework combines environmental awareness, motion planning, and whole-body control to help humanoid robots navigate complex environments while maintaining balance and achieving their goals.
Recently, Galbot’s humanoid robot completed over 100 rallies in an autonomous live tennis game by tracking balls, adjusting positions and returning various shots.
Humanoids overcome obstacles
The new research framework allows humanoid robots to navigate crowded environments by adapting their whole-body movements to obstacles. The system allows robots to choose appropriate movements, such as climbing over objects, ducking under overhead barriers, or turning sideways to get through tight gaps.
The framework addresses a key challenge in robotics: coordinating a robot’s movements with its understanding of the space around it. Humanoid robots have to consider more than just whether their feet can fit through an opening. Your heads, shoulders, arms, and torso must also stay clear of surrounding objects. A robot may have a clear path for its legs, but still collide with a low ceiling or narrow door.
PASSAGE combines environmental awareness, movement planning and whole body control to address this problem. According to the project page, the system uses a planner to generate short sequences of movements and a controller that executes them while reacting to the surrounding geometry.
The framework integrates motion selection and execution into a single planner-controller approach. Instead of relying on separate guidelines for different obstacle types, movements can be combined to overcome different spatial constraints.
The research team collected 100 hours of human motion data across 1,500 virtual scenes. Operators demonstrated overcoming obstacles using virtual reality and inertial motion capture, providing examples of how people adapt their bodies to move through confined spaces.
The system uses this data to link human movements to the environmental conditions in which they occur. During robot operation, LiDAR provides information about the environment and helps the frame take into account obstacles and available clearance.
According to project documentation, PASSAGE runs on an Nvidia Jetson AGX Orin computer and requires no pre-built maps or offboard calculations. This setup allows the system to process environmental information and coordinate movements on the robot itself.
Trials reveal challenges
The research reports a success rate of 98.67 percent in 750 simulated attempts in 150 previously unknown scenes. This number measures whether the robot reached its goal, not whether it avoided every obstacle.
When success required completing the task without touching obstacles, the reported rate dropped to 70.3 percent. The difference highlights the challenge of coordinating a humanoid’s entire body in a confined space.
The team also tested PASSAGE on a Unitree G1 humanoid robot in 50 previously unknown physical layouts. The researchers reported that the robot completed all 50 trials without falling or requiring safety interventions. It passed through 45 of the facilities without encountering any obstacles. These results come from a limited research evaluation and do not demonstrate reliability in all real-world environments.
PASSAGE focuses on a capability that humanoid robots may need to work in furnished rooms, factories and other spaces designed for humans. Adapting posture to the available space could help robots navigate environments where fixed walking patterns are not sufficient.
However, the published results also show that reaching a destination and traversing it without contact remain different challenges. On the project page, the framework’s code is listed as coming soon, so its wider adoption and performance in additional settings remains to be evaluated.
