CV
Curriculum vitae
Contact Information
| Name | Yash Turkar |
| [email protected] | |
| Phone | +1 (609)-302-0396 |
| Website | https://yash.turkar.in |
| yash-turkar-334168ab |
Research Interests
- My research centers on utility-aware robot perception: understanding how the quality of visual data shapes downstream performance, and how robots can act to acquire more informative measurements. I am particularly interested in the perception-action loop for field robots in challenging environments, with an emphasis on active sensing, low-light vision, and robust visual navigation under real-world sensing constraints.
Research Experience
-
Jan 2022 - Present Buffalo, NY
Graduate Research Assistant
University at Buffalo
- Conducted research in perception & autonomy for field robots - VLM-based view planning, active low-light imaging, and mapping-quality metrics. Formulated hypotheses, designed controlled experiments and ablations, built reproducible evaluation pipelines, and validated results across simulation and field deployments.
-
Jan 2026 - May 2026 San Mateo, CA
Autonomy Intern
Skydio
- Developed and deployed learning-based perception algorithms for distribution pole detection and localization as part of the inspection and mapping team.
-
May 2023 - Aug 2023 Houston, TX
Intern - Autonomous Robotics
SLB
- Built a sensing-modality evaluation framework (LiDAR/RGB/thermal), validated on quadrupeds, yielding data-driven sensor/configuration recommendations for autonomous inspection.
-
May 2022 - Aug 2022 Houston, TX
Intern - Autonomous Robotics
SLB
- Benchmarked dense SLAM with mixed sim/field datasets for onshore inspection, improving robustness and frequency of inspection, deployed on quadruped robots in a proof-of-concept.
Teaching Experience
-
Aug 2024 - Dec 2024 Instructor/Teaching Fellow, CSE 4/568 Robotics Algorithms
- Completely re-designed syllabus; created ROS-based programming assignments for 50+ students.
- Taught first-principles autonomy. Material included kinematics, probabilistic algorithms for localization and mapping, planning, and navigation.
-
Jan 2022 - Dec 2023 Teaching Assistant, CSE 4/568 Robotics Algorithms
- Led weekly discussions; held office hours; graded exams.
- Designed auto-graded assignments; maintained Piazza and course GitHub.
Education
-
2022 - Present PhD
University at Buffalo
Computer Science
-
2021 - 2023 Master's
University at Buffalo
Computer Science
-
2016 - 2020 Bachelor's
University of Mumbai
Computer Engineering
Selected Publications
-
2025 Active Illumination Control in Low-Light Environments
Y. Turkar, Y. Kim, and K. Dantu
International Symposium on Experimental Robotics (ISER) 2025, Santa Fe, NM, USA. Preprint arxiv:2506.06394.
-
2026 QAL: A Loss for Recall-Precision Balance in 3D Reconstruction
P. Meshram, Y. Turkar, K. Singh, P. R. Masilamani, C. Adhivarahan, and K. Dantu
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2026. Preprint at arXiv:2511.17824, 2026.
-
2026 Learning Visual Information Utility with PIXER
Y. Turkar, T. Chase Jr, C. Aluckal, and K. Dantu
23rd Conference on Robots and Vision (CRV), 2026. Preprint at arXiv:2409.13151, 2026.
-
2026 Generating Synthetic Illumination Variation from Sparse Co-Located Captures
Y. Turkar and K. Dantu
3rd Workshop on Synthetic Data for Computer Vision (SynData4CV), CVPR, 2026
-
2026 Empir3D: A Framework for Multi-Dimensional Point Cloud Assessment
Y. Turkar, P. Meshram, C. Aluckal, C. Adhivarahan, and K. Dantu
23rd Conference on Robots and Vision (CRV), 2026. arXiv preprint arXiv:2306.03660, 2026.
-
2026 Lightning: Adaptive Illumination Control for Robot Perception
Y. Turkar, S. Sadeghi, and K. Dantu
(Under Review), 2026.
-
2025 TERA: A Simulation Environment for Terrain Excavation Robot Autonomy
C. Aluckal, R. V. K. Lal, S. Courtney, Y. Turkar, Y. Dighe, Y. Kim, J. Gemerek, and K. Dantu
2025 IEEE International Conference on Simulation, Modeling, and Programming for Autonomous Robots (SIMPAR), IEEE, 2025, pp. 1-6.
-
2024 VRF: Vehicle Road-Side Point Cloud Fusion
K. N. Khan, A. Khalid, Y. Turkar, K. Dantu, and F. Ahmad
Proceedings of the 22nd Annual International Conference on Mobile Systems, Applications and Services (MobiSys), 2024, pp. 547-560.
-
2023 Kinematics-Only Differential Flatness Based Trajectory Tracking for Autonomous Racing
Y. Dighe, Y. Kim, S. Rajguru, Y. Turkar, T. Singh, and K. Dantu
2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2023, pp. 1629-1636.
Projects
-
Adaptive Illumination Control for Robot Perception
- Developed Lightning, a closed-loop illumination-control framework for robot perception that improves visual SLAM robustness in low-light and high-dynamic-range scenes.
- Introduced a Co-Located Illumination Decomposition (CLID) relighting model to separate ambient illumination from light contribution and synthesize physically consistent multi-intensity training data.
- Formulated an offline Optimal Intensity Schedule and distilled it into a real-time imitation-learned controller that selects discrete onboard light levels while balancing SLAM utility, power consumption, and temporal smoothness.
-
NightHawk: Active Illumination Control [ISER 25']
- Active illumination and exposure control using Bayesian optimization and a feature-utility metric to boost feature detection and matching in low light.
- Implemented the event-triggered optimization loop, camera/light control, and Mfeat; field tests showed 47-197% gains in tracking/matching.
- Designed a ROS2-controlled 50W LED flood light for active illumination in low-light environments.
-
Language-in-the-loop Culvert Inspection on the Erie Canal
- Language-in-the-loop inspection pairing a VLM with viewpoint planning on a legged robot to inspect a 66m culvert and capture targeted close-ups.
- Integrated VLM prompting, stereo-scale recovery, and a culvert-aware planner on Spot; designed the expert study showing ROI agreement over 80%.
- Designed a low-cost 3D printed pan-tilt camera gimbal controllable using ROS2.
-
PIXER: Learning Visual Information Utility
- Lightweight model that predicts pixel-wise “featureness” and uncertainty to filter keypoints and improve visual odometry with fewer features.
- Designed the Bayesian learning pipeline and evaluation, reducing VO RMSE by ~31% with ~49% fewer features on Davis/KITTI-style data.
-
Empir3D: Multi-Dimensional Point Cloud Quality Assessment
- Four-axis point-cloud quality framework (resolution, accuracy, coverage, artifacts) addressing limits of single-distance measures for 3D tasks.
- Implemented metrics and analysis tools and ran SLAM/completion case studies, showing complementary value beyond Chamfer/Hausdorff.
-
EARTH: Excavation Autonomy with Resilient Traversability and Handling
- Autonomy framework for excavators integrating perception, planning, and hydraulic control, supported by the TERA simulator.
- Led perception and simulation components and integrated safe control/state-estimation modules.
-
Hex-Wife: Hex-rotor UAV Platform
- Low-cost hex-rotor UAV built for AUVSI SUAS 2019, used for autonomous flight tests and payload experiments in mapping/inspection.
- Airframe and avionics design; flight-stack and state-estimation integration; controller tuning; field validation.
Skills
Programming Languages: Python (OpenCV, Open3D, NumPy, Pandas, Torch, TensorFlow, PyTorch3D), CUDA, Java, SQL, C, C++, JAX
Robotics & Simulation: ROS, ROS2, Gazebo, NVIDIA Omniverse Isaac, Gibson-Env, Unity, PX4
Tools & Frameworks: MATLAB, CAD, LaTeX, Linux, Git, Docker, LXC, Rerun
Reviewing
- IEEE International Conference on Robotics and Automation, ICRA (2024-Present)
- IEEE International Conference on Intelligent Robots and Systems, IROS (2025-Present)
- IEEE Robotics and Automation Letters, RA-L (2024-Present)
Honors & Awards
-
2022 CSE Graduate Leadership Award
University at Buffalo
-
2019 Just Joe Sportsmanship Award
AUVSI SUAS
Invited Talks
Autonomous Robotics for Archaeology: IMPERO Summer School, Grosseto, Italy (Remote), June 2024
Excavation Autonomy with Resilient Traversability and Handling: MOOG-UB Workshop, University at Buffalo, NY, December 2024