Dhruv Krishna Email & Phone Number

Bengaluru, Ka, In

Jodhpur, Rajasthan, India

Brief note on my research work under Prof. Suril Shah:Featureless Visual Servoing of Tumbling Objects• Created a dataset of 600k videos of tumbling objects on blender and calculated optical flow• Extracted static features of tumbling object from the calculated optical flow using Convolutional Neural Networks• Performed Position Based Visual Servoing by using the extracted features of the tumbling objectQuality Biased Incremental RRT for optimal motion planning:• Biased the nodes of Rapidly Exploring Tree for better and faster solution trajectories using Deep Reinforcement Learning• Introduced goal bias as a hyperparameter for better results

Jodhpur, Rajasthan, India

Outline of the project pursuedAutonomous Navigation of Mobile Robot:• Mapped the environment through Microsoft KINECT Sensor using Real Time Appearance Based Mapping (RTAB-Map)• Navigated the Pioneer-3 DX Mobile robot in the mapped environment autonomously

Singapore

During my internship, I studied various state of the art pose estimation networks like DenseFusion, HybridPose and Object-Posenet. The study included :1. Understanding the working of the various parts of the network and building the networks with correct specifications.2. Comparing their performance on the Linemod and Occluded Linemod datasets3. Analysis of the computational requirements of the networks by comparing the time taken by them for prediction with and without the use of GPUs. 4. Fine-tuning the network for improving the performance.During my internship I also participated in the weekly reading groups that discussed state of the art developments in the field of Computer Vision, Robotics and Deep Learning. These discussions familiarized me with the various diverse concepts and techniques like Lie Groups, Capsule Networks, Multi-modal Sensory Fusion etc. that were used in various ways to solve problems like Robot Reasoning and Robot Imagination.I also presented a paper to the reading group titled "Making Sense of Vision and Touch: Self-Supervised Learning of Multi-modal Representations for Contact-Rich Tasks" in which, a compact multi-modal sensory representation of complementary visual and haptic data was used to improve the efficiency of performing contact rich task of peg insertion.

Thiruvananthapuram, Kerala, India

Currently, I am working on Vision based Manipulation and Grasping for a dual-arm Half Humanoid robot to achieve tasks like switch panel operations inside a crew module, replacement of CO2 Canisters and performing hand gestures during communication

Mumbai, Maharashtra, India

Parkinson Tremor Management:• Created a light weight and cheaper solution to suppress Parkinson Tremors for the project of Parkinson Tremor Suppression hosted by BETiC Lab, IIT Bombay• The prototype secured first position among the top 21 institutions for technical education (IITs) in India