IKBT is finally online!!

For the past year, I've been developing this package to solve closed-from Inverse Kinematics with Behavior Tree (IKBT).  Inverse kinematics is a hard problem even for human experts. It's difficult to automate because of the high-level mathematical reasoning required. There were few people/groups attempted to do this, but they shared some common problems, such as their algorithms only works on robots with specific DOF and specific configurations.

IKBT, on the other hand, does not have any prior assumptions of the robot. In theory, IKBT works on robots with any DOF (<7) or configuration. What we did differently is that we used Behavior Tree, commonly used in game AI, to build our expert system. We encode the knowledge that engineers often use to solve inverse kinematics as "solver leaves" in Behavior Tree. While solver leaves carry out specific task of identification and solving, higher level parents composite nodes decide how and when to choose a solver leaf. With a few knowledge-based solvers, IKBT can solve most of test robot (18 out of 19), including complex 6DOF robots. The results are proven to be correct through verification.

We're finally at the stage that IKBT can make a public appearance!!

Manuscript:

https://arxiv.org/abs/1711.05412

Code:

https://github.com/uw-biorobotics/IKBT