iansseijelly / depth_obstacle_avoidance

Supplement materials for Micro Quadcopter Obstacle Avoidance with a Lightweight Monocular Depth Network

depth_avoider_crazyflie

Supplement materials for paper: "Nano Quadcopter Obstacle Avoidance with a Lightweight Monocular Depth Network" submitted to IFAC World Congress 2023.

Experiment Video

https://youtu.be/ss8BPzg_JyY

The above video is attached to give reviewers a clearer view of 1. flight trajectories and 2. real-time onboard camera images with predicted depth maps in the paper.

The included experiments:

• With depth network trained in the CyberZoo environment:
  1. Evaluated in the CyberZoo sparse / dense environments with fixed obstacles;
  2. Transferred to the CyberZoo environment with dynamic / unseen obstacles;
  3. Transferred to the Corridor environment.
• With depth network trained & evaluated in the Corridor environment.

Source Code

• NanoDepth.py contains the nano depth convolutional neural network framework written in PyTorch.
• bsm.py contains the behavior state machine based on depth map for obstacle avoidance.