Sarthak Talwadkar

About

Robotics software engineer/researcher focused on perception, SLAM, and sensor fusion, with hands-on experience taking systems from simulation to embedded/real-time deployment. Led transparent-surface (glass) detection using GDNet and achieved a major real-time speedup (~7–9 FPS to ~30 FPS) while preserving >90% recall, and has built ROS-based EKF GPS-IMU fusion plus profiled/optimized Visual SLAM for performance and memory stability. Also brings production-style deployment skills via Docker/Kubernetes orchestration of ML inference services with autoscaling and model update rollouts.

Experience

Education

Key Strengths

• Improved glass detection pipeline performance from ~7–9 FPS to ~30 FPS while maintaining >90% recall
• Profiling-driven optimization (identified backbone as ~60% of inference time; targeted bottlenecks)
• Built GPS-IMU EKF fusion with asynchronous updates (100Hz IMU / 1Hz GPS) using timestamp buffering and interpolation
• Debugged and optimized Visual SLAM performance and stability (reduced ORB features, added caching, CUDA matching)
• Resolved SLAM memory crashes via keyframe database growth control and keyframe culling
• Designed distributed multi-robot communication with latency/failure handling (multi-master ROS, UDP multicast + TCP sync, graceful degradation)
• Simulation-to-deployment workflow: Gazebo worlds/noise models plus containerized, Kubernetes-orchestrated inference services with autoscaling