Topological Mapping
December 15, 2015
Topological mapping is about creating graphical map of the environment using only images. The research challenge lies in detecting loop closures accurately. We propose a method called Bag-of-word-pairs (BoWP) to improve the loop closure detection.
February 19, 2017
In this video, the points to be reached is provided by PTAM. Self-tuning PD controller, which uses gradient-descent to update controller parameters, tries to minimize the error between its current location and the desired location.
February 24, 2017
Formation control is required for carrying out surveillance or monitoring in an environment which is dynamic. Some examples - forest fires, oil spillage, flood situations or even the failure of drones. The fleet or swarm may change its formation in the run time to meet the objectives of the mission. The video below shows some experiments that we have done in our lab.
We are just showing off the working of a PD controller in tracking an AR marker. It needs to be improved further.
September 2014
We developed occlusion reasoning scheme that improves the tracking of multiple targets in a scene. The work was published in AVSS 2015. The paper could be downloaded from this link.
June 2018
Visual Servoing is about guiding a robot motion (computing joint velocities) directly using image features. Challenge lies in selecting suitable image features for which Image Jacobian could be computed analytically. The following video demonstrates the working of a conventional visual servoing where the centroid of the ball being tracked used for driving the robot joints.
December 2013
Video shows capabilities such as Human following, gesture-based control, tele-presence, speech interface, human skin colour detection.
March 05, 2018
In this work, we propose a new deep network architecture to obtain state-of-the-art result for depth and pose estimation from monocular images. The architecture makes use of both spatial and temporal consistency along with a new Charbonnier penalty function to obtain higher accuracy.
September 2017
In this work, we try to learn control policies for Landing manoeuvre using Least Square Policy iteration (LSPI) based Reinforcement learning. The initial movement towards the landing site is done using PTAM based localization and Aruco Marker. Since it is possible to turn on both the cameras (front and down) at the same time in Parrot AR drone, we turn it in sequence. The drone moves upward to get bring the landing marker into its view.
Publication at IROS 2018