The Batmobile Projects


Right Photo:  MERIT 2003:  Patrick Knapp (Syracuse University)  and Jesse Clarke (University of Maryland) - MERIT Fair Best Project Award - RITE Category

Left Photo:  MERIT 2004:  Gaurav Singhal (Columbia University) and Marshall Miller (University of Maryland) - MERIT Fair Best Project Award - RITE Category

The Batmobile projects are two efforts in our laboratory to explore neurally-plausible algorithms for the control of steering based on bat-like echolocation.  Unlike many sonar approaches in the mobile robotics literature, most bats do not appear to estimate echo azimuth based on the time of flight between two widely-spaced microphones.   This is probably due to its small head and thus the very short time of flight between the two ears (max time of flight ~ 70us).  Instead, azimuth is estimated from the relative intensities of an echo at the two ears.   At the ultrasonic frequencies that bats use, the head creates an angle-dependent attenuation.

While one can buy Polaroid sonar ranging modules to report the closest object that reflects a sonar ping, much more information is available from the echoes from all objects in the near field.  The sonar we are working with uses a single transmitter and two receivers which have directional-sensitivity curves that allow intensity comparisons between the two "ears".

In the MERIT 2003 project, information about an obstacle's range and azimuth were combined using a torque summation model where obstacles effectively "nudged" the robot to turn away.  This approach was inspired by work by Bill Warren's laboratory at Brown University on human walking patterns.  While fairly effective as a basic reflex-like maneuver, many aspects of the approach did not produce desirable effects.  The classic reflexive robot trap is to send it into the corner of the room.   Nonetheless, we were able to demonstrate some relatively rapid decision making based on multiple obstacles.

- movie:  DSCN2825.AVI
- movie:  DSCN2834.AVI
- movie:  DSCN2837.AVI
- movie:  DSCN2839.AVI
- movie: DSCN2841.AVI
- movie:  batmobile_2_003.avi
- movie:  batmobile_2_005.avi

In the MERIT 2004 project, we explored a new approach that we called 'Openspace'.  The openspace algorithm was a 'hypothesis' testing approach which evaluated the merit of each possible direction, reducing the desirability of certain directions when an obstacle was detected.  The best direction following this evaluation was selected and the robot turned towards this direction.  This algorithm was successful in three main ways: 1) the robot could select directions to turn towards, even if it meant steering initially towards another obstacle, 2) the robot tends to steer towards the center of an opening between two obstacles, and 3) goal information and obstacle information can be combined in a common data representation.  We have tested this new algorithm on a Koala mobile robot (K-Team, Switzerland).  None of the built-in sensors on the Koala were used.

openspace diagram

    - movie: DSCN4049_s.avi  - Koala robot cuts across the forest of 'trees' (3 MB)
    - movie: DSCN4050_s.avi -  Koala robot steer through a corridor of 'trees'  (3.5 MB)
    - movie: DSCN4052_s.avi -  Steering around a tree trunk (3.9 MB)
    - movie: DSCN4057_s.avi - Off road travel (6.1 MB)

In MATLAB simulations, we have been exploring many other aspects of the algorithm and testing new concepts of merging information about goals and obstacles, including swarms of bats.

- movie : MATLAB simulation of a bat flying through a forest. - os6_05.avi  (3.3 MB)

- movie : MATLAB simulation of a bat flying through a forest with variable repetition rate. - os7_06.avi  (2.1 MB)


- movie : MATLAB simulation of 8 bats : each has a tendency to fly behind another - mbat06_07.avi  (2.3 MB)