Homing in robotics can be defined as the problem of finding a route that enables a robot to return to its initial, "home" position in space, after the execution of a random path. Odometry information has been traditionally used for the solution of this problem but it has been proven to be an unreliable source. In this work, a novel method for visual homing based on a panoramic camera is proposed. We have developed a navigational scheme that enables a robot to move between two positions provided that at least three features of the environment can be corresponded between these two positions.The method relies on the computation of the bearing angle of the mentioned points and does not require the computation of any structure information. Since the correspondence problem has not yet been solved efficiently for images taken from considerably different viewpoints, a tracking scheme has been used to support long-range navigation. More specifically, the KLT corner tracker has been employed to track corners as the robot departs from its initial position. The proposed method selects intermediate positions in the initial path and uses them as intermediate target positions. Then, the robot moves between consecutive intermediate positions according to the bearing angle of the tracked features. The method has been implemented on LEFKOS, a mobile robotic platform of ICS/FORTH. The first experimental results show that homing can be supported with remarkable accuracy, which is not affected by the distance traveled by the robot. Figure 1 describes the adopted strategy.

Fig 1(a): The robot can move between two adjacent
positions S and T given the correspondence of three distinct features in a panoramic field of view.

Fig 1(b): Tracking of corners from start to end positions, can lead the robot from the end position to the (potentially distant) home position, through a number of intermediate positions.

• Contributions: Antonis Argyros, Costas Bekris.
• Current status: Method developed, first on-line experiments from an implementation on lefkos are extremely promising.
• Sample results

Fig 2: Lefkos, the robotic platform of CVRL with
a panoramic camera mounted on it. It is used for experiments on robot homing.

Fig 4: What the robot sees at the initial (home) position and the detected corners. You may also download a video (3,39 Mbytes) showing how this view evolves over time.

*** You may also see the full experiment: Either Observer's view (5,6 Mbytes) or Robot's view (3,39 Mbytes) with corners tracked

Available documents:

ICS-FORTH Technical Report No 287, March 2001.