Rating is available when the video has been rented.
This feature is not available right now. Please try again later.
Published on Feb 1, 2016
We want a robot’s ability to detect objects (e.g., pedestrians, cars, etc.) in its specific operating environment to evolve and improve over time. Our Experience Based Classification framework for robot perception builds on the well established practice of performing hard negative mining when training an object detector. Rather than stopping mining for data once a detector is trained, we continuously seek to learn from the detector’s mistakes during operation. This learning process is entirely self-supervised, facilitated by the fact that we have additional scene context at our disposal in robotics.