Quadruped Robot Masters Agile Terrain at Just 8 Hz Control Speed

A new study challenges the conventional wisdom that faster motion control always makes for better robot locomotion. Scientists have successfully trained an ANYmal C quadruped to move dynamically across rough ground using motion commands updated at just 8 Hz—a far cry from the 200 Hz frequencies typical in industrial robotics.

The Approach#

The team employed deep reinforcement learning to train motion control policies at frequencies spanning from 5 Hz up to 200 Hz. By systematically comparing performance across this range, they uncovered a counterintuitive finding: slower control loops can deliver equal or superior results on real robots.

The trained controller enabled the quadruped to travel at 1.5 m/s across uneven terrain while staying stable when hit by unexpected external forces. These results were achieved on a physical robot, not in simulation.

Why Low-Frequency Control Works Better#

Low-frequency policies appear to be inherently more resilient to real-world robot challenges. The learned controllers proved less vulnerable to:

• Actuation latencies — delays between command and execution
• System dynamics variations — differences between expected and actual robot behavior

So robust was the low-frequency approach that the researchers could perform sim-to-real transfer without traditional techniques like dynamics randomization or explicit actuation modeling—steps usually required to bridge simulation and reality.

Implications for Robotics#

These findings suggest a path toward simpler, more practical robot deployment. By relaxing the need for ultrafast control loops, engineers may reduce computational demands and improve overall system robustness. The authors have released training and deployment code along with detailed analysis to support reproducibility and further research in the field.