Company Description
EdgeVerse India Private Limited specializes in developing next-generation perception stacks designed to operate independently of specific edge devices, with a primary focus on applications for mobility & industrial automation. Our mission is to redefine the capabilities of perception technologies, ensuring versatility and precision across various use cases. Located in Bengaluru, EdgeVerse India offers a collaborative environment for individuals passionate about advancing perception technologies.
The Role
We are looking for a Radar Perception Engineer who can own the classical and learned perception stack from pointcloud to tracked objects. You will design and implement clustering, data association, and multi-object tracking algorithms, and benchmark them rigorously on real-world data. This is a hands-on engineering role; you will write production-quality code, not just research prototypes.
What You Will Do
- Design and implement pointcloud clustering algorithms on mmWave radar pointcloud data
- Build multi-object tracking pipelines - state estimation, data association, track management
- Implement and tune data association algorithms
- Develop and maintain Kalman Filter variants (KF, EKF, UKF) for object state estimation
- Benchmark tracking performance using standard metrics
- Contribute to dataset collection, annotation pipelines, and simulation frameworks
What We're Looking For
Must Have
- Common sense and ability to code and maintain large codebase without the help of AI tools
- B.Tech with 3-4 years or M.Tech with 2-3 years of hands-on experience in object tracking and multi-sensor perception
- Strong fundamentals in estimation theory - Kalman filtering, Bayesian inference, probabilistic data association
- Experience implementing clustering algorithms on sparse, noisy radar pointcloud data
- Solid understanding of multi-object tracking - track lifecycle management (initiation, tentative, confirmed, deleted states), association cost matrices, gating strategies
- Hands-on experience with data association algorithms - GNN, JPDA, MHT
- Kalman Filter family - Linear KF, EKF, UKF, CKF etc.
- Ability to read, explain and implement algorithms from academic papers
Advanced Tracking - Good to Have
- Neural IMM — learning model transition probabilities from data rather than hand-tuning them
- Extended object tracking methodology
- Various motion models
- Particle Filters & Sequential Monte Carlo Methods
- ML & Deep Learning Approaches
Who You Are
- You are comfortable with mathematical rigour - understand, implement and explain to others
- You debug algorithms by understanding the physics and the math, not just blindly tuning hyperparameters
- You can go from a paper to a working Python implementation quickly
- You take ownership: if the tracking is broken, it's your problem until it's fixed - go deep as much as you can until the problem is root-caused and solved
- You know when classical methods are sufficient and when learned approaches are justified - you don't reach for deep learning by default
