SVR-1: Vision based Autonomous Collision Avoidance

Abstract - In this project we developed an algorithm for real-time vision based autonomous navigation for mobile ground robots in an unstructured static environment. The obstacle detection is based on Canny edge detection and a suite of algorithms for extracting the location of all obstacles in robots current field of view. In order to avoid obstacles we designed a reasoning process that successively builds an environment representation using the location of the detected obstacles. The optimal decisions to avoid the obstacles are made in respect to this environment representation.

Tests During The Application Development:

Application Main Modules:

1. Wi-fi PC-Robot Communication Module - This module provides communication support for robot remote control over a wireless TCP/IP connection.
2. Obstacle Detection Module - This module provides the necessary support for obstacle(s) visual identification and localization.
   
3. Decision Agent Module - This module incorporates the robot reasoning process, which successively builds an environment representation and makes optimal decisions to avoid all obstacles on the robots travel path towards a specified goal.

Demo: Coming Soon...
Results: Coming Soon...

NXT vs. SRV-1

Abstract.This paper presents a comparison of performance for the vision sensors and processing power of two, widely used, robotic systems: NXT Mindstorms and SRV-1 Blackfin. The performance analysis was done in relation to an obstacle avoidance algorithm implemented on both platforms. Three case studies were performed: in the first case study we analyzed the accuracy of object recognitions, in the second case we analyzed the performance of multiple object detection and in third case determined the average time it took each robot avoid the obstacle. These case studies show how much the performance depreciates with the use of low level hardware, like the NXT, for the task of obstacle avoidance.

Hybrid Robot

Abstract— The paper presents an algorithm for autonomous robots with the ability to avoid obstacles when traveling between two points on a given trajectory. The algorithm was designed to obtain good performance on simple hardware systems, with low performance sensors. We implemented and tested the algorithm on the NXT kit that incorporates a basic ultrasonic sensor and an NXT camera. Experiments were performed to evaluate the robot performance. Results show that the robot system successfuly avoids obstacles, achieving good performance in terms of accuracy, total time needed to detect the obstacle, process the information, determine the new path for avoiding the obstacle and then returning on the initial trajectory. The algorithm was implemented in Java.

Autonomous mobile robot avoiding obstacles using ultrasonic sensor or video camera

Abstract - This paper presents two mobile robots that have the ability to avoid obstacles when going from a start point to an end point. The robots were built using the NXT Mindstorms kit: one uses an ultrasonic standard sensor (US) and the other a custom vision subsystem MindSensors NXT camera. The software was written in two languages: NXC (NXC/NBC1.7 firmware) and Java (Lejos0.7 firmware). The main algorithm, implemented both with ultrasonic sensor and camera, computes the optimal trajectory for avoiding the obstacle. Experiments were performed to evaluate the robots performance.

Test during development:

Matlab demo results (for (560,421) final point):

A. Travel path
(Click image to enlarge!)


B. Time/distance traveled
(Click image to enlarge!)