Multi-Robot Exploration Using Potential Field

A continuation of James' Multi-robot Fast Map Construction.

Our goal is , building on James' project, to improve the potential field navigation and mapping abilities of the program so as to run it in larger real-world environments.

Resources

Current code found in the RobotExploration repository on BitBucket under the branch edited.

Attached on this page are tests of the program in simulation.

Current work

• Fine-tune vortexing code that moves robot around rather than directly away from obstacles.
  • Vortexing code is added and shows results; but is still inconsistent. Needs to be working on 2Room map before trying Obstacle map again?
• Implement SLAM algorithm to improve robot mapping. (This may not be testable in MobileSim -- might have to use the physical robots?)

Changes made to original code

• Gate-sensing function added to potential attraction calculation.
  • Previously, the robot was not drawn to narrow openings such as doorways or halls due to their being surrounded by an obstacle. With the gate-sensing function, the robot is slightly more attracted to such narrow openings, called "gates", making it more likely to investigate rooms and corridors.
  • Gates show as yellow cells on the maps, though like other cells they stop being attractive once the area around them has been explored enough.
• Limited turning per timestep.
  • Previously, the robot often turned as much as 180 degrees at once, which could cause erratic behavior on simulation but would be harmful to a physical robot. Now, the maximum amount a robot will turn per timestep is defined as a variable in RobotMap.h, making the robot's turns smoother.
• Bugs fixed:
  • All robots now shut down after map is completed, where previously only the one that first received the complete map did.
  • Robot no longer considers itself an obstacle when calculation repulsion by other robots -- this had caused some strange behavior in cell just below robot.
• Small configurations that improve robot's performance in simulation:
  • Attract area spans entire map (should probably change if ever exploring a large physical area)
  • Maximum potential level reduced (less difference between explored and unexplored cells means that robot is less likely to be stuck in a high-potential minimum)
  • Map dimensions increased -- allows for more precise mapping

<meta name="robots" content="noindex" />

-- (c) Fordham University Robotics and Computer Vision