Global Dynamic Window Approach

Oliver Brock

To get a quick impression...

Please take a look at this video(30MB). It is a large file and will take a while to download, but it's worth it!

Other pages:    Elastic Strips    Decomposition-based Motion Planning    Homepage

Introduction to the Global Dynamic Window Approach

Mobile robots mostly refer to robots moving in the plane with three degrees of freedom, two translational and one rotational degree of freedom. There are various approaches in the robotics literature to generate motion for these kind of robots. It is a much simpler problem than for robots with more degrees of freedom, such as mobile manipulators, because the configuration space is low-dimensional. A holonomic mobile robot is a special kind of mobile robot; it is able to accelerate into any direction in the plane at any time. This is different from non-holonomic robots, like cars, for example, which cannot move sideways.

The most widespread approach to generate motion for mobile robots are potential field approaches. These approaches, however, cannot guarantee for the robot to reach the goal position. Other approaches perform global planning on a world representation or map to determine a motion. But in dynamic environments the map is not static; it changes as obstacles move. And in many cases the map representing the surroundings is not known, or changes so frequently that it is hard to keep track of the changes.

The global dynamic window approach provides a powerful motion primitive that does not rely on a map. It explores the environment and generates goal-directed motion on the fly, based upon the information acquired thus far. By incorporating dynamic properties of the robot into the motion generation process, this approach is capable of moving the mobile base at high velocities while avoiding collisions, even with moving obstacles. The global dynamic window approach was tested experimentally with excellent results.

Experiments were conducted with the XR4000 by Nomadic Technologies, depicted here. Unfortunately for the robotics community, Nomadic was acquired by 3COM and will not be building robots any more. The XR4000 is the best mobile base ever built. It is a holonomic mobile base with a caster-based drive system. It is equipped with two rings of sonar sensors, infra-red sensors, bumpers, up to two Pentium-based PCs, a SICK laser range finder (which was used for sensing in these experiments), and much more. It has unsurpassed velocity and acceleration capabilities. Using a dynamically-decoupled controller developed by Bob Holmberg, who also developed the drive system, it executes motion commands with unparalleled precision. A great robot!

A series of experiments are shown in this video (30MB). It is a large file to download, but the results are worth seeing. The robot moves autonomously over distances of over 40 meters in a completely unknown environment at velocities of over 1 m/s. That is twice as fast as any other mobile robot I've seen. The video is not sped up! The motion in every segment was generated by simply specifying a relative position to the robot as the goal position. The robot starts out without any knowledge of the environment and explores the world until the goal is reached. If you read the papers below you will be surprised how simple the algorithm is and how complex the motion behavior that emerges!

During the video you will see a map in the lower left corner of the screen, showing the robot's current knowledge about the world. Obstacles are indicated in black; they are added as the robot perceives them while moving around. A gray gradient is also shown. This gradient is used by the robot to find its way to the goal. It is generated using a wave front expansion algorithm to compute a local minima-free potential function in real-time. The robots path is traced by a line. Various segments of the video display the range of capabilities of the global dynamic window approach: in one segment the robot discovers that its original path was blocked and it has to backtrack. In another segment the robot swiftly and successfully circumnavigates people in a tight space and successfully reaches its goal.


Last modified on November 13, 2001 by Oliver Brock
Visitors since November 8, 2001: provided by