What is this... fun... you speak of?Originally Posted by Xevel
Tician's approach would be so much more fun though
It was mostly just lazy, since some of the math for current 'state of the art' biped gaits makes my head really not happy. So why waste more time with my crapped-out organic computing unit trying to create a natural looking system when I can just assemble a massively complex and poorly constrained system, then do a heuristic search for something that might work. Thinking to add that extra helping of complexity to the system by using a 7+ point spline for every servos' goal angle profile. The feedback/fitness function will largely make or break the idea.
Still not sure of specifics, but rough process: populate angle profiles -> fix profiles using physical angle limits -> penalize profiles using gait angle limits -> fix/penalize profiles using physical/gait velocity limits -> approximate joint torques at multiple instances in full-step-cycle using IMU data and angle profiles -> tweak/penalize profiles using torque limits and torque feedback/calculations -> calculate 'energy' usage of generated solution and also compare resulting trajectory, step sizes, and step frequency to the inputs. Controller needs any two of three inputs: <x,y,z,w> trajectory, left/right <x,y,z> step sizes, and step frequency.
Sorry for this bit of dumping in your thread, but won't be worth its own thread for a while yet.
Reply With Quote
Bookmarks