In some motion planning problems, you might not want just any valid path between your start and goal states. You might be interested in the shortest path, or perhaps the path that steers the farthest away from obstacles. In these cases you're looking for an optimal path: a path which satisfies your constraints (connects start and goal states without collisions) and also optimizes some path quality metric. Path length and path clearance are two examples of path quality metrics. Motion planners which attempt to optimize path quality metrics are known as optimizing planners.
In order to perform optimal planning, you need two things:
- A path quality metric, or optimization objective.
- An optimizing motion planner
You can specify a path quality metric using the
ompl::base::OptimizationObjective class. As for the optimizing planner, OMPL currently provides two optimizing planners that guarantee asymptotic optimality of returned solutions:
You can find out more about asymptotic optimality in motion planning by checking out this paper by Karaman and Frazzoli. The following planners also support
ompl::base::OptimizationObjective, but do not provide theoretical guarantees on solution optimality:
Lastly, the following planners provide a theoretical guarantee of asymptotic near optimality, but currently do not support user-specified objectives:
CForest parallelization framework
OMPL includes support for the CForest parallelization framework, boosting the convergence of optimal planners. Learn more about CForest.
OMPL comes with several predefined optimization objectives:
- path length
- minimum path clearance
- general state cost integral
- mechanical work
Each of these optimization objectives defines a notion of the cost of a path. In OMPL we define cost as a value accrued over a motion in the robot's configuration space. In geometric planning, a motion is fully defined by a start state and an end state. By default, these objectives attempt to minimize the path cost, but this behavior can be customized. We also assume that the cost of an entire path can be broken down into an accumulation of the costs of the smaller motions that make up the path; the method of accumulation (e.g. summation, multiplication, etc.) can be customized.
OMPL also provides users with the ability to combine objectives for multi-objective optimal planning problems by using the
To maximize computational efficiency, OMPL assumes that the cost of a path can be represented with one
double value. In many problems one value will suffice to fully define the optimization objective, even if we're working with a multi-objective problem. In these cases, the cost of a path can be represented as a weighted sum of the costs of the path under each of the individual objectives which make up the multi-objective.
However, there are problems that cannot be exactly represented with this assumption. For instance, OMPL cannot represent a multi-objective which combines the following two objectives:
- minimizing path length
- maximizing minimum clearance
The reason why these two objectives cannot be combined is because we need more than one value to perform the accumulation of the path cost. We need one value to hold the accumulation of length along the path, and another value to hold the the minimum clearance value encountered so far in the path. Therefore, the multi-objective problems that cannot be represented in OMPL are those where the individual objectives do not share a cost accumulation function. We note that the above objective can be approximated by combining a path length objective with a state cost integral objective where state cost is a function of clearance.
OMPL currently does not support control planning with general optimization objectives.
Wanna learn more?
Check out these tutorials:
and the optimal planning demo.