Developing end-to-end hardware-software robotics and autonomy platforms.

I'm Vedant Parnaik — a Robotics AI Engineer with 6+ years building real-time perception, sensor fusion, and SLAM systems. Currently AI Engineer #3 on the founding team at IDrive Inc., shipping end-to-end AI pipelines for Full Self-Driving and ADAS. I'm also building the Follow-Me Rover, a personal robot that follows you and carries your bags, and raising a first check for it.

See the Follow-Me Rover View projects
Vedant Parnaik
01

About

I work at the intersection of deep learning, computer vision, and real-time embedded systems. My focus is making robots perceive the world robustly — fusing cameras, LiDAR, radar, IMU, and GNSS into reliable state estimates that downstream planners and controllers can actually trust.

At IDrive Inc. I'm building the perception backbone for an FSD platform: multi-modal sensor fusion, spatio-temporal motion prediction, and TensorRT-optimized inference pipelines running on NVIDIA Jetson AGX and DGX Spark.

Outside of work I'm building NervLynx, a lightweight robot runtime for Raspberry Pi-class boards. Behaviour is plain Python: try it in a simulator, then run the same project on the robot, with motor safety and a phone dashboard built in. I'm also building Handme Bot, a teleoperated mobile manipulator with a 5-finger bionic hand.

I hold an MS in Robotics from University at Buffalo, SUNY, where I led stereo vision and visual odometry research on the KITTI dataset.

03

Experience

Robotics AI Engineer Founding Team · #3

IDrive Inc.
Los Angeles, USA · Jan 2025 — Present
  • Architecting end-to-end AI pipelines for Full Self-Driving (FSD) and ADAS — multi-modal perception, sensor fusion, and spatio-temporal motion prediction.
  • Designed perception systems integrating Camera, IMU, GNSS, LiDAR, Radar, and telemetry with SLAM and visual odometry for robust localization and object tracking.
  • Trained deep learning models (PyTorch / TensorFlow) for object detection, semantic segmentation, and motion prediction from multi-modal streams.
  • Processed 6 TB of raw sensor data into ML-ready datasets; built large-scale preprocessing, telemetry ingestion, anomaly detection, and closed-loop validation pipelines.
  • Optimized real-time inference with CUDA, TensorRT, and ONNX on NVIDIA Jetson AGX and DGX Spark for embedded automotive platforms.
PyTorchTensorRTCUDAROS2SLAMSensor FusionJetson

Research Volunteer — Image Analysis

University at Buffalo, SUNY
Buffalo, NY · Aug 2024 — Jan 2025 · Advisor: Dr. Mingchen Gao
  • ML research focused on data preprocessing, feature extraction, and augmentation pipelines in PyTorch.
PyTorchMedical ImagingAugmentation

Graduate Student Researcher

University at Buffalo, SUNY
Buffalo, NY · Feb 2023 — May 2024
  • Led stereo vision and visual odometry research on the KITTI dataset — depth estimation, scene segmentation, and 3D scene understanding.
  • Achieved 2-meter accuracy through calibration and preprocessing pipelines for object detection and tracking.
  • Built a custom training framework supporting diverse datasets and model architectures.
Stereo VisionKITTIVisual OdometryOpenCV

Robotics Software Engineer

IndiaFIRST® Robotics
Pune, India · Jun 2021 — Jul 2022
  • Developed an MVP in C++ and Python for real-time sensor fusion and perception analytics on Automated Guided Vehicles.
  • Built data pipelines for ingestion, transformation, storage, and error handling.
  • Prototyped EvoBOT — a 1/4-scale self-balancing robot capable of object grasping and slope traversal.
C++PythonAGVSensor Fusion

Robotics Engineer — Team Rudra (Robocon)

Marathwada Mitra Mandal College of Engineering
Pune, India · Aug 2018 — May 2020
  • Developed autonomous mobile robot software in C++ — navigation, real-time sensor processing, collision avoidance.
  • International robotics competition team — Secured Rank 1 (intermediate) and Rank 11 (final) in India.
C++EmbeddedMobile Robotics
04

Technical Stack

Languages

  • C++
  • Python
  • C
  • C#
  • Bash
  • SQL
  • Assembly

AI & Perception

  • PyTorch
  • TensorFlow
  • Sensor Fusion
  • Motion Prediction
  • Object Detection
  • Semantic Segmentation
  • Depth Estimation
  • SLAM & VO

Computer Vision

  • OpenCV
  • Stereo Vision
  • Sensor Calibration
  • 3D Scene Understanding
  • KITTI / nuScenes

Edge AI & Deployment

  • NVIDIA Jetson AGX
  • DGX Spark
  • TensorRT
  • CUDA
  • ONNX
  • Quantization
  • Real-time Inference

Robotics Frameworks

  • ROS / ROS2
  • Motion Planning
  • Kinematics
  • Visual Odometry
  • SLAM Stack

Systems & Tooling

  • Linux
  • Docker
  • GStreamer
  • TCP/IP, UDP
  • Ethernet
  • CAD Modelling
05

Featured Projects

Now building · Pre-seed

Follow-Me Rover

A personal rover that follows you and carries your bags.

Locks onto one person and follows at about 2.6 m on a 4-motor mecanum base. Re-identification after occlusion and obstacle avoidance pass in simulation against a model of the real drivetrain. The brain, simulator and ROS 2 package are open source as followme-bot.

Raspberry PiYOLO + ByteTrackRe-IDROS 2Python
Pitch deck Repository

Handme Bot

Teleoperated mobile manipulator with bionic hand.

4-wheeled omnidirectional mobile base + 5-finger bionic hand. Real-time kinematics, PWM motor control, and gantry-style end-effector coordination on Raspberry Pi. Extending with a 6-DOF arm + linear actuator for pick-and-place tasks.

Raspberry PiEmbeddedMotion Planning
In Progress
Open source

NervLynx

Lightweight robot runtime for Raspberry Pi. No ROS required.

Write a robot's behaviour as plain Python functions, try it in a simulator on a laptop, then run the same project on the robot. Motor safety, a phone dashboard, and a report of every run are built in. Includes a follow-me template, an optional ROS 2 bridge, and fleet deploys.

PythonRaspberry PiSimulationSafety
Repository

Smart Dashcam Annotator

Auto-labeling tool for dashcam footage.

Python tool that ingests dashcam video and auto-generates annotations for downstream perception model training.

PythonOpenCVAnnotation
Repository

Stereo Vision & Depth — UTV

Image segmentation + depth estimation for off-road UTVs.

Stereo-camera based depth estimation and semantic segmentation pipeline for off-road utility-terrain vehicles — built on KITTI methodology with custom field data.

Stereo VisionPyTorchSegmentationKITTI
Repository

ASL Fingerspelling Recognition

Real-time American Sign Language fingerspelling recognizer.

End-to-end deep learning system for recognizing ASL fingerspelling from video — preprocessing, model training, and inference notebooks.

JupyterDeep LearningCV
Repository

EvoBOT — Self-Balancing Robot

1/4-scale self-balancing prototype.

Prototype 1/4-scale self-balancing robot capable of object grasping and stable traversal on slopes and rough terrain. Built during my time at IndiaFIRST Robotics.

C++Control SystemsEmbedded
Archived
See all repositories on GitHub
06

Publications

Earlier manuscript · 2026

NervLynx: A Contract-First Runtime Framework for Reliable and Observable Robotics Pipelines

Vedant Parnaik

Earlier write-up of NervLynx as a contract-first pipeline runtime. The project has since become a lightweight Raspberry Pi runtime: Python behaviour, simulation, motor safety, and a phone dashboard.

View on GitHub
Earlier Work

Survey Paper on Smart Cultivation and Prediction System for Agriculture

Vedant Parnaik et al.

Earlier Work

Constructive Reuse of Polymers into Plastic Paver Blocks

Vedant Parnaik et al.

07

Let's build something

I'm building the Follow-Me Rover and raising a first check. If you invest in robotics, want to run a pilot, or are building something similar, I'd like to talk. Also happy to chat about NervLynx, embodied AI, or anything at the intersection of code and the physical world.

Email
vedantparnaik@gmail.com
LinkedIn
in/vedantparnaik
GitHub
@vedantparnaik
Phone
+1 (716) 486-8097
Send me an email Read the pitch deck