CV

Education, experience, projects, publications, certifications, and technical skills.

Contact Information

Name Shubham Derhgawen
Professional Title Research Assistant, Motion Planning Lab, UC Riverside
Email sderh001@ucr.edu
Phone +1 (951)-465-9836
Location Riverside, California

Professional Summary

Robotics researcher and software engineer (M.S. Robotics, UC Riverside) working on simulation, evaluation, and learning for robot navigation around people. First author of CrowdROS, under review at ICRA 2027.

Experience

  • 2025 -

    Riverside, California, US

    Research Assistant
    Motion Planning Lab, UC Riverside
    Developing CrowdROS, a ROS 2 and Unity platform for developing and testing autonomous robot navigation in crowds (first-author paper under review).
    • Wrote ROS 2 nodes in C++ with custom messages, QoS profiles, and TF2 frames; handled simulation-time sync between Unity and ROS 2.
    • Built environment maps with slam_toolbox and integrated simulated LiDAR, RGB camera, and IMU streams.
    • Trained CADRL navigation policies with reinforcement learning and exported them to ONNX for inference inside ROS 2.
    • Benchmarked local planners (CADRL, DRL-VO, Arena) on safety, path-quality, acceleration, and jerk metrics; ran legacy planners in Docker to isolate dependencies.
    • Automated unattended batch runs of 75-episode suites with fixed seeds and automatic pass/fail (success, collision, timeout).
    • Built telemetry logging and rosbag replay tools, and captured failure cases as config-driven scenarios re-run against each planner.
    • Integrated crowd policies with motion matching and smoothing, reducing jerk by 9% and angular jerk by 21%.
    • Wrote pytest tests for the benchmarking metrics and the documentation other lab members use to run their experiments.
  • 2025 - 2025

    Riverside, California, US

    Grader, Reinforcement Learning (CS258/EE227)
    University of California, Riverside
    • Graded coursework for CS258/EE227: Introduction to Reinforcement Learning, covering MDPs, TD learning, and policy gradient methods.
  • 2025 - 2025

    Riverside, California, US

    Visiting Research Student
    Collaborative Intelligence Systems Lab, UC Riverside
    Built infrastructure for human-in-the-loop autonomous driving experiments and multimodal data collection.
    • Built a 3-vehicle, 3-machine CARLA platform with Logitech G923 wheels for human-in-the-loop driving experiments.
    • Ran a modified CARLA build in Docker that streams camera output as video instead of individual images.
    • Developed a record-and-playback pipeline for real-time 720p simulation and high-resolution multimodal capture; recorded runs to MCAP and reviewed them in Foxglove.
    • Mounted and wired vehicle sensors and operated sessions for real-world multi-agent driving data collection.
  • 2022 - 2024

    India

    Software Developer
    Vijeta Transport Private Limited
    • Developed and maintained a production transport management web system using Node.js and Firebase.
    • Built real-time data handling and Power BI dashboards giving visibility into the fleet, billing, and cashflow.
    • Worked directly with operators and stakeholders to turn their requirements into deployed features; automated browser testing with Selenium.
  • 2020 - 2022

    India

    Systems Engineer
    Tata Consultancy Services
    • Built and maintained production data pipelines in C# and SQL on Azure Data Factory, Data Lake, and Batch.
    • Built Azure DevOps CI/CD pipelines with automated test and deployment stages for C# ETL workflows.
    • Supported migration of on-premises ETL workflows to Azure, and monitored, debugged, and optimized production jobs to reduce downtime.

Education

  • 2024 - 2025

    Riverside, California, US

    Master of Science
    University of California, Riverside
    Robotics
    • Advanced Robotics
    • Reinforcement Learning
    • Foundations of Machine Learning
    • Computational Methods for Robotics
    • Artificial Intelligence
    • Foundations of Robotics
    • Information Theory
    • Advanced Computer Vision
    • Natural Language Processing
    • Focus area: Embodied AI and Machine Learning.
  • 2016 - 2020

    Pune, India

    Bachelor of Engineering
    University of Pune
    Information Technology
    • Fundamentals of Data Structures and Algorithms
    • Database Management Systems
    • Operating Systems
    • First Class with Distinction.

Acknowledgments

  • 2026
    Acknowledged Contribution: Drive My Way
    Collaborative Intelligence Systems Lab, UC Riverside

    Acknowledged for integrating Logitech G-series steering and pedals and supporting driving data recording (non-author contribution).

  • 2026
    Acknowledged Contribution: CooperScene
    Collaborative Intelligence Systems Lab, UC Riverside

    Acknowledged for hands-on data collection, including driving the car during collection sessions (non-author contribution).

Publications

Skills

Programming Languages: C++, Python, C, C#, JavaScript, SQL, Bash
Autonomy: State machines, Nav2, MoveIt, RRT*, A*, SLAM (slam_toolbox, GMapping), Kalman filter, PID, LQR, FK/IK
Robotics Software: ROS 2 Jazzy, ROS, TF2, URDF, QoS, rosbag2/MCAP, RViz, Foxglove, Docker, Git, Linux, pytest, CI/CD
Simulation: Gazebo, Unity, CARLA, PyBullet, highway-env, Isaac Lab (introductory), scenario and regression runs
Learning and Vision: PyTorch, reinforcement learning (PPO, VPG, CADRL), imitation learning (BC, DAPG), DPO, Transformers, ONNX, Optuna, TensorBoard, Faster R-CNN, COLMAP, OpenCV
Hardware: TurtleBot3, Dynamixel and hobby servos, PCA9685, Arduino, ESP32, Raspberry Pi, I2C, UART, PWM, SolidWorks, 3D printing
Cloud and Web: Azure (Data Factory, Data Lake, Batch, DevOps), AWS (EC2, RDS, S3), Node.js, Firebase, Power BI, Selenium

Certificates

  • Azure Data Scientist Associate - Microsoft
  • Azure Fundamentals - Microsoft
  • Accessibility in Action - Microsoft
  • Problem Solving (Basic) - HackerRank

Projects

  • Scene and Social Awareness in Trajectory Prediction Dec 2025

    Self-proposed final project for EE 267, with Aadhar Chauhan.

    • Trained a trajectory-selection policy with Direct Preference Optimization over kinematic rollouts in highway-env.
    • Ranked candidates on safety, lane keeping, and progress, cutting obstacle violations by 47% before vs. after training.
  • 12-DOF Quadruped Robot Jun 2025

    Team of 3, with Shubham Jangle and Soumya Mondal.

    • Built a 12-DOF legged robot in a 3-person team: assembled it and helped with the SolidWorks design and wiring.
    • Wrote Arduino C++ firmware driving 12 servos via a PCA9685 driver over I2C, commanded over WiFi through an HTTP server.
    • Built a Python Tkinter GUI for joint control, servo calibration, and gait testing, with JSON pose save/reload.
    • Streamed live video from an ESP32-CAM to an OpenCV viewer with an FPS overlay and remote quality control.
    • Debugged across hardware and software with the team: traced Arduino resets under load to voltage sag in the wiring.
  • Computer Vision Pipelines Jun 2025

    Advanced Computer Vision, UC Riverside.

    • Fine-tuned Faster R-CNN (ResNet-50 FPN) in PyTorch with mixed precision to detect pneumonia in 26.7k RSNA chest X-ray studies.
    • Ran COLMAP structure-from-motion and aligned two partial 3D reconstructions; built a video tracker with ResNet-50 appearance matching.
  • Transformers and Vision-Language Models Jun 2025

    Natural Language Processing, UC Riverside.

    • Implemented a Transformer encoder-decoder from scratch in PyTorch and trained it for machine translation.
    • Built ViT patching, a CLIP contrastive loss, LoRA adapters, and REINFORCE-based RLHF in course exercises.
  • State Estimation, Control, and Navigation May 2025

    Foundations of Robotics, UC Riverside.

    • Implemented Kalman-filter state estimation and PID and LQR control for a TurtleBot-style mobile robot.
    • Implemented A* grid planning in ROS, deployed over SSH to a physical, Dynamixel-driven TurtleBot3 Burger.
  • Autonomous Warehouse Pick-and-Deliver Robot Jan 2025

    Self-proposed final project, with Harshini Murugadoss. Simulation only.

    • Built a ROS mobile manipulation pipeline in Gazebo for a TurtleBot3 with an OpenMANIPULATOR-X arm.
    • Mapped the warehouse with GMapping (2D LiDAR SLAM) and inflated the occupancy grid by the robot radius for safe planning.
    • Implemented RRT* from scratch on the inflated grid and a closed-loop waypoint follower for collision-free navigation.
  • Sample-Efficient RL with Expert Data Augmentation Dec 2024

    Self-proposed final project for CS 258.

    • Implemented PPO from scratch in PyTorch with GAE, clipped policy updates, and entropy regularization.
    • Added Behavioral Cloning and a DAPG demonstration loss, reaching 257 ± 39 on LunarLander-v2 in 1,024 environment steps (single seed).
    • Built an Optuna and TensorBoard framework for hyperparameter optimization and multi-seed benchmarking.
  • Hopper Legged Locomotion Dec 2024

    Reinforcement Learning, UC Riverside.

    • Implemented VPG and PPO from scratch (clipping, GAE) to train one-legged hopping in PyBullet.