70 likes | 189 Views
16-735 Project Progress Presentation Coffee delivery mission. Nov, 13, 2007 NSH 3211 Hyun Soo Park, Iacopo Gentilini. Landmark detection. Changed the robot design to get better resolution (camera closer to the mirror): Used three landmark detection in “good regions”. 2. L 1. L 3. 3.
E N D
16-735 Project Progress PresentationCoffee delivery mission Nov, 13, 2007 NSH 3211 Hyun Soo Park, Iacopo Gentilini
Landmark detection • Changed the robot design to get better resolution (camera closer to the mirror): • Used three landmark detection in “good regions” 2 L1 L3 3 1 L2
FLAN algorithm • Extraction of landmark features in the image: • m_lmk : RGB mean of landmark region; • s_lmk :RGB standard deviation; • d_lmk : Euclidean distance of a pixel when its RGB is different from m_lmk about 1.96 · s_lmk . • Calculating landmark regions • d : calculate Euclidean distance d for each pixel tom_lmk and scaling each component by s_lmk; • if d ≤ d_lmk pixel is in the landmark region and update value to 255 else update value to 0. • Get landmark position • for each pixel = 255 check neigh- borhood: if its dimension is bigger than the threshold r_lan, calculate the center of mass of the neighborhood o position given by built in sensorsx calculated position using three landmark detection
Path planning : D* algorithm - Nomad Simulator : MATLAB
Path planning : D* algorithm - Program on Nomad according to simulator (C++) • Problem • Noisy sonar : filter required • Robot size
Motion algorithm • LOAD map • SET D* backpointer • REPEAT • MOVE to TargetPoint • READ sonar; • state estimation using Kalman filtering; • if robot is in a landmark region, update state vector using landmark information; • UpdateMap and D* • UpdateTargetPoint • IF current position == goal position, THEN return • END REPEAT
Scheduling • Kalman filtering • position level filtering using sonar data and wheel encorder • Combining image processing algorithm