Learning to Walk and Trot with Parametric Gait Generation!
Most other solutions for smaller quadrupeds nearly all implement the 2-beat trot gait, where each pair of two diagonal feet raise and move at the same time. Even though only two feet are supporting the robot during a majority of the cycle, it is not that hard to implement since if you lift and move and drop the feet fast enough, the robot does not have enough time to tip over toward a raised foot.
In the 4-beat "walk" cycle, 3 feet are always on the ground, so it seems simpler. But as it turns out, the 4-beat walk cycle is actually harder to implement because you need to shift the center of gravity significantly with every step. Otherwise, the moment a foot gets raised, the entire robot begins to tilt and fall down toward the corner of the raised foot. Even if you raise and lower the foot pretty fast, the motion will be so jerky that it simply won't work. By shifting the CoG as each foot gets raised in turn, the robot can maintain balance between the three legs on the ground.
The moment this gets implemented correctly, the walking motion of the robot suddenly starts looking very lifelike. Imagine a tiger slowly walking along, shoulders rising and dropping as it moves.
EvoDog can now walk, trot, turn or sidestep with desired speed control at runtime, with all the parametric gait code squeezed into an Arduino Nano. Now on to integrating an IMU without increasing memory usage. The goal is to innovate as much as possible on the control and hardware to allow bringing the cost down to levels not previously thought possible, while providing excellent performance.
Comments