TrajOpt

TrajOpt, short for Trajectory Optimization, is an open-source framework for planning smooth, collision-free trajectories for robotic systems. It is designed to compute trajectories that satisfy kinematic and collision constraints while optimizing criteria such as smoothness and energy usage. The framework is used in robotics research and industry for manipulation tasks with robotic arms and mobile robots.

Approach: TrajOpt formulates the plan as an optimization problem over a discretized trajectory, parameterized by a sequence of waypoints or joint configurations over time. The objective combines terms for trajectory smoothness (minimizing acceleration or jerk), tracking performance, and sometimes end-effector pose accuracy. Constraints include joint limits, dynamic feasibility, and obstacle avoidance. Collision avoidance is handled through penalties or through locally convexified obstacle constraints, enabling the use of convex optimization techniques. The solver typically employs an iterative procedure, solving a sequence of convex subproblems (sequential convex optimization) to handle non-convexities.

Implementation and usage: TrajOpt is an open-source project with a modular API. It has been used with robotic arms and manipulation tasks and has seen integration with common robotics software stacks such as ROS and MoveIt in various research setups. It supports scripting interfaces and can interchange cost terms and constraints to fit different planning problems.

Limitations and status: As with other optimization-based planners, TrajOpt may be sensitive to initialization and can converge to local minima. Real-time performance depends on problem size and available computing resources. It has spurred subsequent work in trajectory optimization and local convexification methods.