The demo is based on four different steps that are described in this page.

At the very beginning of the demo, the environment is completely unknown. The first step of the demo consists in building a map of the environment in which the robot is going to navigate. This task is semi-autonomously performed by the software SEGLOC developed at LAAS. The robot is moved by an operator in the environment. The points detected by the front laser range finder are gathered into line segments. A 2D-map of the environment is iteratively built by fusing the segments of different views, using Kalman filtering techniques.

Map built by SEGLOC: the 2D line segments are represented by 3D boxes in blue. The yellow zones are forbidden zones for the robot. They are added by the operator. This environment was built in a couple of hours in october 2002 during the SITEF (Salon International de Technologies du Futur) in Toulouse.

In the second step of the demo, the line segments of the map built by segloc are extrapolated into vertical boxes. Additional 3D obstacles can be added to the map. Through an interface, the user can specify a goal position for the robot in the map of the environment. Then the path planning platform Move3D automatically computes a feasible collision-free path between the initial configuration of the robot to the goal. This step needs to take into account the constraints of rolling without slipping induced by the wheels of the robot and of the trailer.

In the following step, the robot starts following the planned trajectory, dealing with:

• inaccuracy of the map of the environment,
• possible localization errors,
• unexpected obstacles that are not in the map.

These perturbations result in the same side effect: the path initially computed by the path planner may be in collision. To make the robot react to these perturbations, we have developed a generic approach of trajectory deformation. This method iteratively adapts the current trajectory in order to make it move away from obstacles, keeping the kinematic constraints of the system satisfied.

For short range motions (< 30 meters), the odometric sensors of the robots are sufficient to perform localization. Even in tight environments, the reactive path deformation method can correct the localization errors. However, for long range motions, odometry is not enough anymore. SEGLOC the map building software also performs global localization with respect to the map of the enviroment. The position of the robot is computed by matching the line segments seen by the laser range finder with the line segments of the map.