How Amazon builds AI-enabled perception for robots
summary
Robin is a package manipulation robot that segments and classifies cluttered conveyor scenes using an instance segmentation model and a defect classification model, handling novel shapes, reflective sensors, deformables and subtle damage. Training moves from a few hundred hand‑annotated images to large-scale models via fine‑tuning, synthetic scenes, data augmentation and automated continuous learning that detects distribution shifts, samples rare defects from the long tail, retrains and redeploys; hundreds of stations run now with thousands planned.
AI-written summary · about
chapters (10)
- 0:00Introduction to Amazon robots
- 1:11Robotic manipulation challenges
- 3:40Types of packages at Amazon
- 6:52Key challenges in robotic manipulation
- 11:52Introducing Robin robot
- 14:23Robin challenges in production
- 17:20Robin perception systems
- 19:51Scaling Robin perception models
- 26:19Continuous learning for scaling
- 27:45Conclusion and future plans
related talks
18mHow Amazon builds and deploys autonomous robots
28mBuild smart robots and operate them easily with AWS Robotics
13mEnhancing Object Detection & Avoidance Systems for Robots in Challenging Environments Using AI
15mRobot Hand Driven by Computer Vision and Reinforcement Learning
29mAmazon Lookout for Vision
04mAI and the Robotics Revolution
17mBeyond Automation: AI and the Evolution of Robotic Intelligence
18mRobotics: why now?