Autonomous forklift

End-to-end solution for autonomous forklift.

Deep learning
Computer vision
CUDA
ConvNets

2017

About project

Task

Develop a self-driving forklift software prototype, capable of safe driving within warehouse without operator from point A to point B.

Solution

  • Driving prediction module based entirely on video input from serveral cameras using conv nets
  • Control module — driving acceleration, braking and rotation angle of steering wheel
  • Data module — datasets collection and augumentation
  • Training module — end-to-end conv net

In a few words

Revolution in the field of machine learning is here and now. Every advance in Deep Learning technologies finds its applicability in various applications. And, of course, in self-driving vehicles too. We are talking not only about end-user applications like self-driving cars, but also about commercial applications like automatic forklifts for warehouses.

Together with a large Russian company a self-driving forklift was created, capable of safe driving within warehouse without operator from point A to point B. An interesting feature — we use end-to-end training, that means the only input for training is real actions of the driver and corresponding video from camera. Potentially, the training set is infinite: there is no manual segmentation and objects detection. Because of that there is a lot of opportunities for rapid deployment in any new warehouses.

Below you can see a few video examples that show how system works on predetermined route.

Video 1. Training process of the model.

Video 2. Training process of the model.
Next project

Annotation tool