Real-Time Freespace Segmentation Using Deep Learning on Autonomous Robots for Detection of Negative Obstacles
CSEF · 2019 Computational Systems & Analysis (Senior Division Only)
Overview
Objectives Many small unmanned ground robots are being developed to perform tasks such as delivery, surveillance, household tasks, and many other formerly-human tasks. It is essential to these robots' core functionality that they are able to navigate difficult terrain, requiring advanced perception capabilities. Although a significant amount of work has been done on the detection of standing obstacles (solid obstructions), almost no work has been done on the detection of negative obstacles such dropoffs, ledges, downward stairs. Detecting these negative obstacles using reliable, cost-effective sensors is crucial for the success of autonomous robots. Methods Small autonomous robot running Robot Operating System and an embedded GPU which was used to run the neural network, along with a desktop GPU used to train the network. Results This research developed a method of terrain safety segmentation using deep convolutional neural networks. The custom semantic segmentation architecture uses a single camera as input and creates a freespace map distinguishing safe terrain and obstacles. The network was trained using heavy data augmentation, enabling the network to generalize well, even when using very small hand-labeled datasets. The results showed that the system generalizes well, achieving around 94.9% mIOU accuracy on the validation dataset. Conclusions The neural network is deployed to an embedded GPU on an indoor robot. Because of its computationally- efficient design, the network is able to run at 55 fps and create a freespace map that can be used to create a costmap for navigation and obstacle avoidance. Experimentation with the neural network combined with pathfinding algorithms proved the robot's ability to reliably detect and navigate around both standing and negative obstacles in real-time, using only an RGB camera and the neural network developed in this research.
Summary statement
I developed a novel method of terrain safety segmentation on autonomous robots using deep neural networks.
Help received
None. I developed and trained my neural network, and tested it on the robot, completely by myself.
Competition history
- CSEF 2019
Resources
Related projects
ISEF · 2019
Real-Time Freespace Segmentation Using Deep Learning on Autonomous Robots for Detection of Negative Obstacles
CSEF · 2018
Deep Learning Based Collision Avoidance Algorithm for Mobile Robots in Pedestrian Environments
CSEF · 2017
Autonomous Off-Road Vehicle Using Computer Vision for Surveillance Applications
ISEF · 2021
Effective Object Detection Neural Network on an Autonomous Robotics Platform Applied on TPU and Other Systems
CSEF · 2026
Real-Time Disaster Search and Rescue System Utilizing Semantic Segmentation and Object Recognition onboard Fixed-Wing UA
CSEF · 2016
Autonomous Robot Navigation Using Computer Vision for Exhaustive Path-Finding
ISEF · 2018
Deep Learning Real-Time Object Detection Through Convolutional Neural Networks Using OpenCV and Optical Flow Algorithms for the Visually-Impaired
CSEF · 2017
Safecopter: Developing a Collision Avoidance System Based on an Array of Time-of-Flight 3D Cameras
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects