Jaskaran Singh Sodhi

Dec '23

Jul '23

Apr '23

Mar '23

Mar '23

Feb '23

Nov '22

Jul '22

May '22

Mar '22

Sep '21

Sep '21

Mar '21

July '19

Developed an end-to-end solution for subcanopy wildfire monitoring, in a GPS-denied environment.

Skills: Systems Engineering, Multi-Spectral Sensing, Autonomous Exploration

This work explores (i) a self-refinement method to ground VQA respondes to the query image (to reduce hallucinations), (ii) a multimodal routing framework to learn failure patterns in SoTA models, in the realm of common sense reasoning for self-driving.

Skills: VLMs, Self-alignment, Text-image grounding

This project aims to leverage large field-of-view (FOV) cameras to explore point cloud prediction approaches, and essentially develop a 3D perception stack for a multi-camera large FOV setup.

Skills: Multi-view fisheye reconstruction, occupancy prediction

In this work, we explore generative methods to augment sampling-based planners with informed sampling capabilities, when constrained kinematically.

Skills: GANs for informed sampling, learned heuristics for sampling-based planning, kinematically constrained RRTs, Dubins car

We attempt to explore how we can efficiently exploit stereocameras in premapped LiDAR environments to get the best of both worlds - a low-cost relocalization solution in an HD map.

Skills: Photometric residual minimization, heterogenous relocalization, Ceres Solver

Worked with Shubham Shrivastava and Abhyuday Puri on perception for off-road driving.

Developed and productionized a multimodal multi-task model for ego trajectory prediction, from driver imitation.

Worked under the supervision of Dr. Sebastian Scherer at AirLab, CMU on subcanopy wildfire monitoring solutions.[website]

Also researching deep optical flow for fisheye odometry.

Worked under the supervision of Dr. Joydeep Biswas at AMRL, UT Austin on learning-based navigation for cluttered environments.

My work focused on developing learned heuristics for informed sampling of sampling-based kinematic planners, for deploying in cluttered environments.[blog]

Worked under the supervision of Mr Siddharth Jha.

Worked on the feature matching pipeline for UAV-based offline 3D reconstruction.

Worked under the supervision of Mr Haneesh Allu and Mr Besta Prem Sai.

Worked on path planning and obstacles avoidance modules for autonomous UAVs for warehouse inpection.

Working under the supervision of Dr. Debashish Chakravarty at AGV, IIT Kharagpur on Stereo-camera Localisation in Mapped Environments [blog].

We present a novel probabilistic planner and an integrated MPC controls stack for efficient multiple waypoint navigation.

Co-organised by IIT Kharagpur and DRDO, India, as part of the Inter IIT Tech Meet 10.0, this challenge aimed to develop a UAV-guided navigation system which could guide cars on snowy terrains.

Our team won first place, and also secured the Inter IIT Tech Meet 10.0 overall General Championship.

The task was to develop (in simulation) a mobile robot with a manipulator capable of multiple waypoint navigation and exploration. As the winning solution, we created a probabilistic route planner, which finds near-optimal solutions while respecting the computational capabilities of the robot and integrated this with an adaptive MPC.

We presented this work at the International Conference on Control and Robotics 2022 in our paper "Multiple Waypoint Navigation in Unknown Indoor Environments"

Member of the IIT Kharagpur - IUPUI, Indiana - USB Colombia collaborative team.

Designed tightly/loosely coupled high-speed localisation in for racecar localisation in pre-mapped LiDAR circuit. The localisation was using 3 static-state LiDARs. [code]

Developed a ROS-based sensor testing module as part of the Base Vehicle Software team. [code]

Co-organised by IIT Guwahati and DRDO, India, as part of the Inter IIT Tech Meet 9.0, this challenge aimed to develop a purely-vision based UAV navigation and exploration system, and ArUco marker landing for flood relief and delivery.

Our team won first place, owing to our modified gbplanner solution.

Team Lead of the Mars Rover Team of the Autonomous Ground Vehicle Research Group, IIT Kharagpur

Developed the wheel, chassis and suspension for rover prototype with 15 deg gradeability and max speed 20cm/s.

Designed a 5-DOF modular robotic manipulator with 2-finger grip for semi-autonomous on-board equipment repair.

A open-source Python package to demonstrated the difference in performance between sampling-based planning algorithms such as RRT, RRT* and Informed RRT*.

An open-source implementation of graph-planning for 2R manipulators.