trajopt: Trajectory Optimization for Motion Planning

trajopt is a software framework for generating robot trajectories by local optimization.

The following core capabilities are included:

• a solver for non-convex optimization problems, using sequential convex optimization.
• cost and constraint functions for kinematics and collision avoidance
• constructing problems from JSON-based specification format

The core libraries are implemented in C++ (API docs), and python bindings are generated using boost python.

The theory and technical details of this software are described in a paper.

Source code is available on github.

Contents:

• Installing trajopt
  • Dependencies
  • Instructions
  • Common installation problems
• Tutorial
  • Move arm to joint target
  • Move arm to pose target
  • Increasing reliability with multiple initializations
  • Rolling your own costs and constraints in python
  • Optimizing the base position and torso height
  • Walking with humanoid robot
  • Obstacle geometry from sensor data
• Tips and troubleshooting
  • Optimization gets stuck in a local minimum
  • Optimization is slow
  • General troubleshooting tips
• Miscellaneous
  • Downloading test data
  • Collision costs
  • Finding out about available costs and constraints

Comments/questions/problems? Please post them to our discussion group.