New record: 13.2 m/s for a quadruped robot
In experiments on a treadmill, the BlackPanther2 (BP2) robot, weighing 36.5 kg, reached a speed of up to 13.2 meters per second - according to the study's authors, this is a new record for quadruped robots. This result was achieved thanks to a new approach to motion control that takes into account the physical limitations of actuators in dynamic conditions. The presented solution not only exceeded previous limits but also confirmed that there is a real possibility of achieving speeds comparable to fast animals in the context of robotics.
This record is not just a technical achievement - it has significance for future applications. High speed in quadrupedal locomotion opens up new possibilities in logistics, search and rescue, where speed and flexibility are key. However, it should be emphasized that this achievement applies only to one specific robot model and test conditions - it does not automatically mean that all quadrupeds can now move at such a speed.
Why are accurate actuator models key to stability?
Enlarged imageClose zoomPrevious imageA key element of the new approach is an improved actuator model that accounts for physical nonlinearities, such as torque coupling at high speeds and magnetic saturation. Previous methods often neglected these effects during simulation training, leading to a large discrepancy between simulation and actual robot behavior - the so-called sim-to-real gap. The new model allows for a more accurate representation of the actual torque characteristic as a function of speed, which is essential to ensure stability during high-speed movement.
Research has shown that without such modeling, learning algorithms can exploit unphysical strategies - e.g., assigning torques greater than physically possible. This leads to instability and failure in reality. The use of an accurate model prevents this situation, which is a key conclusion for the development of advanced control systems.
Adaptive command scheduling - key to stable training
The cooperation between a precise actuator model and a new approach to training within reinforcement learning is crucial. The proposed method uses a two-stage training strategy and adaptive command scheduling (ACS), which gradually increases the difficulty of motion commands during learning. This ensures that the system is not exposed to instability in the initial stage, when the robot has not yet mastered basic movements.
ACS allows for smooth adjustment of the range of commands - e.g., speed and direction - during training, which prevents interference with the algorithms and improves their ability to generalize. Research has confirmed that this method not only increases stability but also does not negatively affect the final performance - which is a significant achievement in the context of dynamic robotics.
Significance and limitations of the results
Achieving a speed of 13.2 m/s on a treadmill is a significant step forward in the development of quadruped robotics. It shows that the use of accurate physical models and intelligent training strategies can significantly exceed previous limits. This opens the way for more realistic motion systems in difficult terrain, where speed and precision are equally important.
However, these results have significant limitations. The study concerns one specific robot - BlackPanther2 - and test conditions on a treadmill. A speed of 11.65 m/s in open terrain is lower, which shows the influence of external factors. In addition, there is no information about the durability of the actuators under such loads or the possibility of scaling this method to other classes of robots. The conclusions cannot be automatically extended to other systems without additional research.
It is worth emphasizing that achieving a speed of 13.2 m/s does not automatically translate into commercial solutions. First of all, the durability of the actuators under cyclic loading at the maximum parameter level remains unstudied - there is a risk of overheating and degradation of materials in long-term use. In addition, the method based on precise modeling of actuators requires significant computing power for real-time adaptation, which may limit its application in small or energy-efficient systems. The conclusions from the study are also limited to one type of robot and test conditions - it is not yet possible to talk about a universal method for the entire class of quadrupeds.
However, this step towards more accurate physical modeling opens up new possibilities for future projects that will be able to use a similar approach to develop dynamic and stable motion in difficult terrain. From a robotics perspective, this is not only a speed record, but also proof that understanding the physical limitations can be the key to transforming simulations into real-world action.
