Abhinav Bhamidipati

IR

Inception RoboticsSep 2024 – Mar 2026 · College Park, Maryland · 3 roles

MedSort – Vision Pipeline for Medication Sorting

• Developed a real-time computer vision system for automated medication sorting using a 3-axis vacuum end-effector; deployed OCR to extract lot numbers, expiry dates, and medicine labels for structured data creation.
• Tuned preprocessing and inference pipelines and optimized CUDA inference kernels for Jetson deployment, enabling low-latency OCR and increasing sorting throughput in pharmacy automation workflows

Terrain-Adaptive UGV Autonomy & Perception Stack

• Designed and deployed full-stack autonomy pipelines for Off-road UGV on embedded Linux platforms (Intel NUC) using C++, Python, and ROS2; integrated multi-sensor fusion (LiDAR Point Cloud + visual odometry) with 3D ICP scan-matching and ORB tracking, improving pose estimation accuracy and system robustness in simulation (Isaac Sim, Unity) and field tests
• Developed reinforcement learning–based terrain-adaptive planning models, achieving 90% navigation success in dynamic environments; automated training & deployment via Docker-based CI/CD (GitLab) & AWS-hosted simulation pipelines.
• Integrated DWA-based local planner with LLM-driven natural language interface to enable goal-directed behavior and reactive obstacle avoidance; validated through SITL pipelines with multi-run regression tests for deployment readiness.
• Adapted a CLIP-(Transformer-based) multi-modal vision-language model for object damage assessment via multi-angle semantic alignment; fine-tuned embeddings and evaluated using retrieval accuracy and anomaly score thresholds.

OR

Computer Vision & SLAM EngineerJun 2024 — Aug 2024 · Onki Robotics, New York City

• Optimized SLAM and localization pipelines for UGVs on Jetson platforms using ROS2, reducing pose update latency by 25%.
• Enhanced environmental perception with sensor-fusion of LiDAR and stereo camera data, generating denser occupancy maps.
• Developed ML-based human feature extraction and tracking models for robust person-following through crowds.