Mini Pupper 2 robot masters gait with AI that handles 320ms delays
Researchers closed the sim-to-real gap on a $50 robot using a time-aware neural network.
Deep Dive
A low-cost Mini Pupper 2 quadruped robot was trained using a time-aware neural network that models actuator delays (>50ms). The approach produces a central pattern generator—a self-sustaining rhythmic gait—robust to +320ms latency perturbations. This biologically inspired method closes the sim-to-real gap for cost-constrained hardware, enabling robust locomotion without expensive sensors.
Key Points
- Mini Pupper 2 experiences >50ms transport delay, making locomotion partially observable
- Time-aware neural network with forward delay model produces a self-sustaining CPG gait
- The learned gait withstands +320ms latency perturbations, closing the sim-to-real gap
Why It Matters
Brings robust, biologically inspired locomotion to sub-$1,000 robots, enabling broader robotics research and education.