Shyam Sreenivasan

MS Robotics @ Northeastern University | Robotics Software Engineer

Building scalable, reliable systems at the intersection of robotics, machine learning, and real-world autonomy.

About Me

I am a robotics graduate student at Northeastern University with over seven years of prior industry experience as a software and machine learning engineer. My background includes designing and operating production-scale distributed systems, real-time data pipelines, and ML platforms for anomaly detection and forecasting.

I'm working at the intersection of AI and robotics, where intelligent systems must operate safely and reliably in the physical world. My interests include motion planning, control theory, simulation-based learning, and the observability infrastructure required to deploy and scale autonomous systems in production environments.

My work spans large-scale software systems, machine learning pipelines, and robotics systems. In industry, I built and operated production ML infrastructure for anomaly detection, forecasting, and observability at scale. In robotics, I focus on motion planning, simulation-driven learning, and real-time control, with an emphasis on reliability and system behavior under edge cases.

Representative Projects

Motion Planning for Ackermann Vehicles (Dubins RRT Variants)

Problem class: Nonholonomic motion planning under curvature constraints

Evaluated nonholonomic motion planners for Ackermann-steered vehicles, focusing on performance trade-offs between feasibility, runtime, and planning success.

  • Approach: Implemented and benchmarked RRT, BiRRT, and RRT* variants with Dubins constraints across 300 simulation trials
  • Outcome: Identified BiRRT*-Dubins as optimal with 10× faster planning (0.043s average) and 100% success rate, while quantifying 292% computational overhead relative to holonomic planning
  • Key learning: Constraint handling dominates performance more than sampling strategy in nonholonomic planning
View Project →

2D Robot Navigation Simulator with LIDAR-Based Obstacle Avoidance

Problem class: Motion planning, Trajectory, Control and Obstacle Avoidance/p>

Created Python-based robot navigation simulator with RRT path planning and adaptive P-controller achieving 90%+ goal-reaching success in obstacle-dense environments through integrated LIDAR sensing, multi-gain control architecture, and real-time collision avoidance.

  • Approach: Developed autonomous navigation system implementing RRT* path planning with hierarchical control architecture (Manual, State Machine, P-Controller) using 360° LIDAR sensing and differential drive kinematics in Python/Pygame
  • Outcome: Achieved 90%+ goal-reaching success through multi-gain controller design (Kp=0.05 steering, Ko=0.02 avoidance, Kv=0.7 deceleration), enabling safe navigation through corridors with adaptive waypoint selection and proximity-weighted obstacle avoidance
  • Key learning: Velocity modulation (Kv gain) combined with reactive steering proved more effective than steering-only approaches, reducing collisions by 75% while maintaining real-time 60 FPS performance
View Project →

Learning from Long-Tail Driving Scenarios in Simulation (Ongoing)

Problem class: Simulation-based learning, reinforcement/imitation learning, rare-event analysis

Building an end-to-end pipeline using the CARLA simulator to surface rare and safety-critical driving scenarios, focusing on identifying long-tail events and improving agent behavior through reinforcement and imitation learning.

  • Approach: End-to-end data collection → scenario mining → evaluation workflows → policy improvement loop
  • Focus: Emphasis on data pipeline design, scenario identification, and evaluation-driven autonomy development rather than purely algorithmic novelty
  • Key learning: Long-tail data challenges and simulation realism limits are the primary bottlenecks in autonomy development
View Project →

Real-Time Kinematics and Trajectory Control for UR5 6-DOF Robot

Problem class: Robot kinematics, control, real-time systems

Implemented a complete kinematic and control framework for a UR5 6-DOF manipulator, replacing an unreliable iterative inverse kinematics solver with a differential velocity-level controller.

  • Approach: Built differential velocity-level controller to enable stable, real-time trajectory execution
  • Outcome: Successfully executed circular, square, and sinusoidal trajectories with sub-3mm tracking accuracy, with interactive visualization of joint states and end-effector pose
  • Key learning: Numerical stability and control-loop design are critical for real-time execution constraints
View Project →

Software Engineering Systems and AI/ML Experience

In prior roles, I built and operated production-scale distributed systems and ML pipelines handling high-volume telemetry and real-time decision-making. While the code is proprietary, the architectural patterns and operational lessons directly inform my robotics and autonomy work.

Skills & Tools

Python C++ Java ROS2 Gazebo OpenCV PyTorch scikit-learn Kafka RabbitMQ Docker Kubernetes AWS MySQL PostgreSQL Prometheus Grafana Motion Planning Robot Control Simulation Real-Time Systems

Contact