On Kalman Filters and Robotics
summary
A working agricultural robot fuses camera-based deep learning weed detections with a Kalman-filtered navigation stack that combines IMU and wheel-encoder inputs; prediction and update alternate in a predict/update loop and the Kalman gain weights uncertainty (example: state uncertainty 400 vs IMU 144 vs encoder 36, giving an updated displacement ~1.3 m). The system runs without a global map, switches a sprayer when the estimated pose is over a weed, and succeeds in-field; failure modes include occluded weeds under cotton and high-speed scenarios that demand much higher sensor FPS.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:35Example of Autonomous Driving
- 2:17Obstacle Avoidance
- 4:04Two Steps to the Kalman Filter
- 7:06Project Management Example
- 16:57What Exactly Is a Kalman Filter
- 19:13Kalman Gain
- 19:30Output of the Kalman Filter
- 25:56Extended Kalman Filter
- 26:42Solution
- 27:24Extended Kalman Filter Can Fail
- 29:44Where Else Extended Kalman Filter Fail
- 30:06Unscented Kalman Filter
- 31:07Nonparametric Filters
- 35:00Application of the Kalman Filter in Agriculture
related talks
49mUsing Deep Learning for Perception in Autonomous Systems: A Perspective from Control Theory
37mLiving with Robots - HRI in the real world
10mUsing Machine Learning to Reduce World Hunger
31mDeep learning for self-driving vehicle technology
27mFrom perception to action: World Foundation Models (WFMs) for autonomous robots and vehicles
33mReducing the environmental impact of agriculture with robotics
15mRobot Hand Driven by Computer Vision and Reinforcement Learning
25mFrom seed to store: AI to optimize the indoor farms of the future