Tarun R

Tarun Ramakrishnan

I am an MS by Research candidate in CSE at IIIT Hyderabad and a graduate researcher at the Robotics Research Center (RRC), advised by Prof. Madhava Krishna and Prof. Sourav Garg.

My research lies at the intersection of 3D computer vision and robotics, with a focus on robot perception algorithms that enable reliable autonomous navigation. Currently, I am focused on visual-inertial SLAM and relocalization.
Last updated: October 28, 2026
tarun.ramak@gmail.com | LinkedIn | GitHub | Google Scholar | Resume

Publications

SILICA: Repurposing Diffusion Priors for Joint Glass Segmentation and Depth Estimation

Tarun R, Anuj Verma, Laksh Nanwani, Sourav Garg, Madhava Krishna
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026
Project Page | Code | arXiv

CrowdSurfer: Sampling Optimization Augmented with Vector-Quantized Variational AutoEncoder for Dense Crowd Navigation

Naman Kumar*, Antareep Singha*, Laksh Nanwani*, Dhruv Potdar, Tarun R, Fatemeh Rastgar, Simon Idoko, Arun Kumar Singh, Madhava Krishna
IEEE International Conference on Robotics and Automation (ICRA), 2025
Project Page | Code | arXiv
* Equal contribution

Projects

Traversability-Aware Indoor Navigation

2026

Used diffusion foundation model priors to segment the traversable floor and estimate depth indoors, and wrote a Numba-based GPU MPPI controller that is soft-constrained to plan trajectories only within the traversable region.

Slides

Transparent Object Grasping Using Modified SILICA

2026

Trained SILICA with CLIP conditioning to predict glass objects and glass scenes separately, greatly reducing false positives. The resulting segmentation and depth feed NVIDIA GraspGen for grasp generation, with an RRT* planner to reach and pick up the object.

Slides

Omnidirectional Vision-LiDAR Calibration and Fusion Framework

2025

Wrote a C++ ROS 2 driver and image stitcher for the Insta360 camera, and designed a synchronization method from scratch to align its untimestamped images with a Livox MID-360 LiDAR. Also built a targetless extrinsic calibration pipeline for the LiDAR, five RealSense cameras and the Insta360 using LightGlue. This setup was used to collect over 2 TB of multi-sensor data in a shopping mall.

Slides

Surfing Scout the Quadruped

2025

Designed a 12-servo controller PCB with micro-ROS firmware in C, and wrote a ROS 2 actuator driver, body-to-leg inverse kinematics, and a C++ IMU-based controller that tracks target body motions.

Slides

Teaching and Service