Slasher is an open 1/10 scale robot car platform for aggressive (up to 70 km/h with stock gearing) deep neural network driving experiments, using neuromorphic silicon retina dynamic vision sensor (DVS) event cameras. It enables E2E dataset collection and localization. E2E dataset collection is enabled via the race controller and first-person-view camera. Localization is via the Loco Positioning System Decawave radios. Control of the human/computer controlled car mode and safety stop is via the BlueTooth joystick remote control.

Slasher was developed by the Sensors Group of the Inst. of Neuroinformatics, Univ. of Zurich and ETH Zurich. See more datasets and tools from the Sensors Group.

If you use Slasher, please cite the accompanying AICAS 2019 paper

Y. Hu, H.M. Chen, T. Delbruck, “Slasher: Stadium racer car for event camera end-to-end learning autonomous driving experiments” AICAS 2019, Mar 2019, Hsinshu City, Taiwan.

bibtex

1. Slasher github project - see readme.md there for detailed software instructions
   1. Slasher mechanical drawings - mechanical plans for slasher body parts
   2. Slasher reference CNN - the reference Resnet-20 steering prediction network
   3. Slasher ROS nodes (Add link to code used for ROS)
2. Slasher datasets - shared via ResilioSync private BitTorrent. Datasets collected with slasher in RPG DVS rosbag format.

Slasher platform

Steering angle prediction on foyer and joggling datasets

Prediction results

Videos

ResNet-20 network architecture

This reference CNN was reported in paper for steering control on foyer and jogging trail tracks