For the past few weeks, I have been implementing a paper titled Simbicon. It is a framework for applying control on bipedal walkers. So far, I have implemented the controllers, provided a GUI for setting the parameters and states for the 7-link biped, and provided an option for pushing the biped forwards or backwards. I have found a stable gait as shown in this video:
The next goals are to find other gaits that I can use to interpolate between/extrapolate from. Finding gaits is a lot of trial and error and intuition. It is difficult process, but hopefully one that can be automated by using a supply of stable gaits and exploration of the state space by the agent.
Some effects that I have noticed from playing around with this:
- For states 2 and 4, setting the angle of the swing hip too far from the center axis (90) causes the biped to pitch forward
- For state 1 and 3, setting the angle of swing hip farther away than around 128 also causes the biped to pitch forward due to moving the center of mass too far from the stance leg. (The stance leg thus has to move farther back to compensate for this)
- Setting the torso pitch backwards (leaning back) allows for the swing hip to go higher.
- It is sensitive to initial conditions though. Results vary between pitching forwards and backwards.
- By leaning the torso backwards, the stance hip will try to come up to keep the biped from falling.
- The stance hip will come up when trying to keep the torso upright.
- When standing by itself on one leg, the biped oscillates in motion. Does this mean to change the gain values?
- There is feet slippage
- If I try to keep the foot parallel to the ground, the biped does a shuffle.
- A foot angle that tries to push against the ground does a slipping motion
- More torque on the ankle joint causes the foot to slip. Torque could come from setting the target ankle angle too far or setting the gains too high