Potential Field Method: Real Robots

Question

The potential field method is a very popular simulation for navigating robots. However, has anyone applied the Potential Field method to real robots? Any link or any requirement to use the method in real robots ?.

Answer 1

I had previously planned field-based planning, but abandoned it in favor of more suitable approaches to my problem. It works well in environments with precise localization and accurate sensor readings, but even more so in real conditions (this is not an excellent solution even in terms of speed and quality of the path, even in simulation). Given that there are currently many good SLAM implementations available either free or cheaper, I would not overestimate if you have no particular problems with reuse. For MRDS (what I'm working on) there is Karto Robotics, ROS has a SLAM implementation, and there are several open source versions for google search only.

If you need a good overview of the various approaches to route planning, you might want to get a copy of the “introduction to autonomous mobile robots” by Segwart et al. Its a pretty good book, and the path planning section gives a good overview of the various strategies.

Answer 2

I would advise you to read the book Planning Algorithms by Stephen M. Laval, if at all interested in planning a path or movement. The methods described in this book are actively used in the robotics community.

A google search for a scientist or IEEE website, on the other hand, will give you many links to documents describing the use and research of a potential field method.

Answer 3

A quick Google for potential field methods brought up this article: Potential field methods and their inherent limitations for Mobile Robot Navigation and reminded me of problems the last time I worked with a potential field method.

In our projects ( CWRU Mobile Robotics ), we saw these exact problems with potential field-based algorithms. Last attempt, a mobile robot to compete on IGVC in 2009, had the same problems described in this paper, especially with local minima and unable to pass through closely located obstacles. I clearly remember that I have to solve problems with closely spaced obstacles, trying to plan through a narrow hole in the fence as part of the navigation task of the IGVC GPS route.

We were able to get a pretty decent planning speed from the algorithm using custom OpenGL shaders to perform all the calculations, presenting the potential field as an image / framebuffer. As Tom points out, this is not so good in unknown or dynamic environments, because in these situations the potential field will never stabilize and will constantly need to be updated.

Answer 4

As @Tom noted above, you cannot usually rely on excellent sensor readings or engines moving you exactly the way you thought you told them.

I had a chance to use a relatively new approach to SLAM many years ago; this is a generalized Voronoi graph ( GVG ); basically, stay at an equal distance from the nearest two walls, keep moving, and at the points where you are equidistant to three or more walls, come back and try each double-walled branch at some point. You will build a schedule that takes you all the way around the room and ensures that you have direct visibility of everything in the room.

Answer 5

We tried and abandoned the potential field algorithm (OpenSteer) for our DARPA Grand Challenge (Team Overbot) in 2003. This is not a good idea for a non-holonomic robot, because it does not take into account steering restrictions or dynamics. It does not work well at all in difficult places. This is best suited for aircraft where you have a lot of open space and you do not want to approach obstacles.