Projects

Roadmap Based Robot Motion Planning in Dynamic Environments

Prog. Lang. Used: Python

Keywords: Dynamic Planning, PRM, State Time Space, A*, KD Tree

Path Planning of mobile robots in an environment which encompasses both dynamic and static obstacles reamains to be a challenging task. Here a pragmatic algorithm based on the roadmap for the static part of the scene is first developed. The roadmap generated is used to form a time-optimal trajectory from start position to goal position without colliding with the non-static obstacle. This is achieved using the PRM algorithm which takes random samples from the configuration space of the robot, tests them for whether they are in the free space, and uses a local planner to attempt to connect these configurations to other nearby configurations. The starting and goal configurations are added in, and a graph search algorithm is applied to the resulting graph to determine a path between the starting and goal configurations.

The final path is obtained in two phases; the first phase is the local level search where a sub-optimal trajectory is developed using the depth-first search (DFS) and in the following phase, an optimal path is obtained taking in consideration time as one of the parameters and using A* search algorithm. The solution obtained is applicable to all types of robots, in any configuration space where the motion of obstacles is unconstrained but should be known beforehand. Furthermore, the velocities of obstacle also should remain constant throughout the process. This approach has been applied to a multi-robot system and is also effective for both free-flying and articulated robots.

Design and Simulation of LQR (Linear Quadratic Regulator) Controller for a gantry crane

Modelling of fruit picking robot

The 3 basic needs of human beings are food, shelter and clothing. Parallel, the world population is also growing at a very steep rate. Hence, natural resources such as land and water is always under continuous stress. Moreover, due to deforestation and global warming cultivable land is decreasing.Hence it is necessary to have effective solutions to iterative tasks like fruit picking, ploughing, sowing etc.So innovation in the agricultural yield will result in great impact on mankind and economy. Articulated robots for fruit picking is a technology that enables autonomy in agriculture processes. As a robotics engineer, I feel it is my moral responsibility to automate arduous jobs done by farmers and increase yield. Hence, this is my motivation to choose the following robot as my modelling project.

So in this project I worked on designing the robot model with its link length and reach. Then its inverse and forward kinamatics was developed. Which was further used in Matlab to do simulation of whole system. So this project involved full design cycle of Robot.

Design of Algorithm for Lane Detection and Turn Prediction used in Self Driving Cars

Prog. Lang. Used: Python, OpenCV

Keywords: Homography, Image Processing, Histogram

The booming topic of self driving car needs no formal introduction. Lane Detection and Turn Prediction being the most fundamental needs of the same. Not RADAR or LIDAR, but it is a front facing camera that can achieve this in the cheapest way. Although we can assume the lanes of the road to be parallel to each other, the challenge is that the actual camera does not look at the road ’from the top’, but it always looks at the road at the same angle.

In this project the homography is first computed using points from an image frame. Image processing techniques such as Edge Detection, noise removal and extraction of ROI is also performed. Perspective transform is taken to build the Histogram of the Lane Pixels for both the lanes, using the homograpy. Further refining of the lane detection is done by fitting a polynomial in between the lane candidates along with turn prediction.

Implementation of Traffic Sign Detection and Classification using MSER and SVM Model

Post the initial image processing, the Detection of the Signs is achieved using the Maximally Stable External Region (MSER) Algorithm, which is a varient of affine trasnformation. It allows multi-scale detection without any smoothing, which can detect both fine and large structures.

The Classifcation phase is implemented using HOG descriptors to train an SVM model with the RBF kernel. Computer Vision techniques is used to get the HOG values. SVM's predict method is then used to check the correctness of the model. The E_out achieved was 97.4% showing that 97% of the images were classified correctly. The code and implementation is show in the links below.

Object Tracking using Lucas Kanade Template Tracker

Prog. Lang. Used: Python

Keywords: Optical Flow, Lucas Kanade, HIstogram Equalization

Optical flow basically is the distribution of apparent velocities of movement of brightness pattern in an image. Sequences of ordered images allow the estimation of motion as either instantaneous image velocities or discrete image displacements. Lucas Kanade is a differential method used for the estimation of the above achieved by combining information from several nearby pixels. Howver this algorithm assumes that the displacement of the image contents between two nearby instants (frames) is small.

In this project, the algorithm has been implemented on three video Visual Tracker benchmark database: featuring a car on the road, a human walking, and a box on a table as shown in the BLOG link. To initialize the tracker a template has been defined by drawing a bounding box around the object to be tracked in the first frame of the video. For each of the subsequent frames the tracker updates an affine transform that warps the current frame so that the template in the first frame is aligned with the warped current frame. The tracked has also been made more robust, by incorporating steps to increase illumination invariance i.e using LAB color space and histogram equalization.