Hi all,

I'm announcing the release of trajopt, a library for trajectory
optimization. More specifically, trajopt is designed for planning
collision-free paths for robot arms and mobile manipulators.

Trajopt is built on top of OpenRAVE. You can define your optimization
problem in Python or C++ in JSON format and then call the optimizer.

Some highlights of trajopt:
- It's fast. It solves arm planning problems in simple environments in
about 150ms (converging to a locally optimal solution)
- It reliably finds collision-free paths, especially with multiple
initializations. FWIW it solves 100% (204/204) of problems in our benchmark
collection
- It performs well on very high-dof problems, e.g. jointly optimizing over
the arms and base of a mobile manipulator, or optimizing over all of the
joints of a humanoid robot.
- A wide variety different costs and constraints are implemented. (Pose
constraints, velocity constraints, static stability, and more.) You can
write your own cost and constraint functions in python or C++.

Source code: https://github.com/joschu/trajopt
Documentation: http://rll.berkeley.edu/trajopt
The technical details are described in a paper, which is linked to on the
front page of the documentation.

This code is at an early stage of development. I'd be grateful to hear
about any problems, questions, or comments.

John Schulman
PhD Candidate, UC Berkeley, EECS Department