Projects
A selection of featured projects showcasing research depth, industry-scale development and a focus on safety, robustness and real-world impact.
Research
Guided Diffusion based Generative model
A research initiative to design a guided diffusion architecture that synthesizes adversarial scenarios for autonomous driving. This work focuses on generating scenarios capable of detecting potential failure modes in perception, planning and control systems, thereby improving the robustness and safety of real-world vehicle deployments.
- Designed a Guided Diffusion Model that generates adversarial scenarios to predict violations in an autonomous driving system
- Synthesized a dataset using Carla with benchmark architectures like: Transfuser, Interfuser, Transfuser++, LMDrive for safe and adversarial scenarios
- Trained the designed Diffusion Model architecture, to create new adverserial scenarios and validate over the existing models
Deep Neural Network
A foundational deep learning project aimed at building and evaluating models for object detection and tracking.
- Creation and training of custom neural architectures from scratch on simulated visual data
- Comparative performance analysis and refinement of baseline models
- Focuses on feature extraction and temporal coherence on image dataset
- More details of the project: Deep Neural Network
Industry
3D Perception and Localization
A perception module exploiting multi-sensor fusion (LiDAR, Radar, Camera) to provide robust object classification, clustering and target tracking, in 3D space.
- Developed the Perception algorithm using Lidar, Radar, Camera system, involving Object clustering and Classification on the 3D point cloud data
- Utilized State estimation and Data association techniques to build a robust Target Tracking algorithm
- Validated with real sensor streams and benchmark datasets
- More details of the project: 3D Perception and Localization
Perception System with Occupancy Grid Mapping
A perception system using front, rear and side sensors to compute occupancy in vehicle surroundings, for enhanced parking solution
- Developed and deployed high-resolution Occupancy Grid Map, using Front & Rear Lidar and Side Radar, to achieve real-time map updates with spatial consistency
- Integrated mapping outputs with control modules for actionable vechicle control
- More details of the project: Occupancy Grid Mapping
Path Prediction and Motion Planning
A motion planning and predictive control algorithm, to estimate motion (for pedestrian/vehicle) and proactively avoid collisions in dynamic environments
- Designed motion predictors that quantify pedestrian trajectory likelihoods
- Developed a path prediction model to estimate collision between pedestrian and vehicle
- Built robust and safe controllers based on the estimated trajectory
- More details of the project: Path Prediction and Motion Planning
Path Planning and Controls
A fuzzy logic–based path planning and motion control system for automated parallel parking, integrating lateral and longitudinal control to achieve accurate single-maneuver parking
- Designed an integrated feedback and feed-forward controller to improve trajectory tracking in constrained parking spaces
- Generated parking trajectories to guide the vehicle safely into the identified space
- Evaluated the proposed fuzzy control against conventional lateral controllers to demonstrate improved maneuver accuracy
- More details of the project: Path Planning and Controls