A* in Action – Artificial Intelligence for Robotics

18 thoughts on “A* in Action – Artificial Intelligence for Robotics”

1. That is so cool.

2. Really cool video! 🙂

3. Freaking awesome <3

4. I understand that A* does the global path planning, but what algorithm is used for control and local planning? DWA?

5. waaaaa

6. How many control primitives you've used for this demonstration?

7. Reference/citation?

8. To me it seems like an RRT planning algorithm rather than A*. Even for global planning -> A* isn't designed for dynamical replanning.

9. It is nicely explained, can explained about simulator?

10. I am writing a similiar simulation for my Master's Thesis and I am wondering how you guys implemented the geometry contstraints for the RRT paths in regards to the movement constraints of the vehicle model. I already have a fully functioning vehicle model and simulation, but in order to use RRT my car needs to be able to move from A to B automatically following its movement constraints. I could probably hardcode it, but surely there must be a nicer way to approach the issue. Anyone has done something similar or knows how to approach the problem? Would be grateful for any suggestions!

11. Magnific#

12. What software did they use?

13. What is the software name? Can anybody help?

14. The creators should do this in 2020 with better hardware it would be interesting to see how it would change over different hardware

15. It seems like either RRT or PRM Path Planning to me.

16. I believe the reason for the A* in the video appears so fast is because they are technically cheating. In the beginning of the video, the environment is too simple, and the A* with the Dubins curves finds the path easily, in milliseconds, which is reasonable since there is not much obstacles. But as the car moves and discover the maze, it needs to replan, but the replanning is take from where the car is to the goal, and this path (from where the car is to the goal) doesn't have much obstacles yet because the car didn't explore it yet. So, I believe that most of the planning the algorithm do is in a environment with few obstacles and that is the reason for the algorithm appears so fast in the video. It is worth mention that the environment, as the car explores, doesn't have a large free space and this decreases the number of possibles states in the exploration trees, decreasing the running time of the algorithm even in the presence of obstacles.
    I am not trying to say that the algorithm is bad or something like that. The algorithm is amazing. I am just explaining in case someone wants to know how the algorithm in the video can be so fast while our implementation isn't (this happened to me).
    But anyways, if there is some way to make the A* as fast as in the video, let me know because I want to know.

17. Seems like RRT*

18. How do i add the notion of history or 'already-visited-room' to path planning?
    I am working on Artificial potential fields tho

Comments are closed.