Although it still requires a driver to operate, ORBITRON is primarily controlled by software. ORBITRON can react to opponents faster than human operators using its state-of-the-art localization and attack avoidance system. It is equipped with 4 stereo pairs of cameras which allow it to map the entire world around it in 3d. Using this map, ORBITRON knows exactly where the opponent is at all times. ORBITRON automatically detects its opponents position and can autonomously coordinate attack sequences with quick and accurate movements.
Below is a demo showing the 3d point cloud constructed from the 4 pairs of stereo cameras. The entire battlebox is mapped in 3d around the robot, which allows orbitron to locate its opponent.
Using a custom 3D simulator, ORBITRON can be tested against human-operated opponents without damaging any real components. The simulator runs ORBITRON’s code exactly as it does in the real robot, including simulating cameras to test computer vision code.
ORBITRON’s attack strategy relies on the fact that it can react faster and more accurately than human drivers. It is designed not to just respond, but to anticipate. ORBITRON outsmarts its competitors by predicting their moves based on their velocity, thereby preempting their actions and issuing a timely counterattack.
ORBITRON doesn't just react — it plans. It uses sophisticated algorithms to extrapolate the opponent's path based on their velocity. By predicting where the adversary is heading, ORBITRON can strategize and intercept the opponent's path. This predictive approach allows ORBITRON to block and corner the adversary, giving them minimal room to maneuver.
The ORBITRON team is exploring new algorithms and designs for vision and localization. Although ORBITRON was proposed to track opponents solely using it’s 4 sets of stereo camera pairs, an additional camera can be attached to a drone and flown overhead. This gives ORBITRON a birds-eye view of the field at all times, as opposed to a 3d view from the ground.
Above is a prototype of the drone localization system. It is much simpler than the 3d tracking pipeline.
The drone camera will be able to extract both ORBITRON’s and the opponent’s position AND orientations at all times. In addition, there are no obstruction problems since the camera is overhead. The drone’s computer will send the tracking data back to the robot directly without having to stream high resolution video, thus minimizing delay.