Ask Your Question

What are the downsides of disabling homotopy class planning in Teb Planner ?

asked 2019-09-16 09:06:31 -0600

robot_new_user gravatar image

According to the documentation of the Teb Planner, enabling homotopy class planning requires much more CPU resources. Disabling this would mean that only one trajectory is optimized. But what exactly do we lose if multiple trajectories are not optimized in parallel in terms of quality of the path planning ? What are advantages of having multiple optimized trajectories and how does the planner make a choice between them?

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted

answered 2021-04-29 07:52:31 -0600

Maxim Rovbo gravatar image

Based on what I have observed, disabling homotopy class planning makes planning much more vulnerable to being stuck with infeasible trajectories when trying to move in narrow spaces or otherwise very close to obstacles. But it does speed up local planning a lot when many obstacles are nearby. Performance tends to drop several times if there are many obstacle cells near the robot with homotopy enabled.

As far as I understand, TEB optimizes a trajectory locally, preferring to modify an already existing plan a bit rather than choosing a completely different one, even when an almost direct trajectory to the goal is available, or the chosen plan is very close to obstacles. When homotopy is enabled, there are several trajectories that are optimized at the same time, so there are more to choose from and the chosen trajectory can jump between very different routes as the result.

edit flag offensive delete link more


To add to this answer:

teb uses NLP (Non-linear Programming) to calculate paths. NLP is an optimization technique that starts from a 'guess solution' and iteratively find better solutions until termination criteria are met. NLP uses a cost function and its gradients to do this. The cost function combines multiple criteria eg. obstacle separation, vehicle constraints and travel time. NLP is local, meaning it converges to the solution nearest to the starting guess. Obstacles cause multiple local minima to exist, therefore the solver could give sub-optimal or infeasible solutions, while better trajectories exist.

That is where the homotopy classes come in: every path in the same homotopy class will converge to the same final result. teb constructs a candidate path in every discovered homotopy class and optimises all of them, selecting the path with the lowest cost function value. Disabling homotopy class planning can lead to suboptimal or infeasible paths.

Vreden gravatar image Vreden  ( 2021-04-30 09:41:48 -0600 )edit

Also, looking at the bottom of the FAQ, the author suggests that 2 homotopy classes might be enough for most cases.

Vreden gravatar image Vreden  ( 2021-04-30 11:30:32 -0600 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools



Asked: 2019-09-16 09:06:31 -0600

Seen: 201 times

Last updated: Apr 29 '21