Autonomous Vehicles
This project explores AI-driven autonomy for vehicles—focusing on perception, decision-making, and safety—to enable safer, more efficient transport across roads and controlled environments.
Project Overview
The Autonomous Vehicles initiative focuses on building and testing AI components that allow vehicles to perceive their environment, plan safe trajectories, and execute control commands—while always prioritizing safety and interpretability.
Current tracks include:
- Perception: Detecting vehicles, pedestrians, lanes, traffic signs, and signals.
- Prediction: Anticipating the motion and intent of nearby agents.
- Planning & Control: Generating collision-free, comfortable paths and following them precisely.
Objectives
- Maximize safety with conservative decision-making and fallback strategies.
- Reduce human interventions across diverse real-world scenarios.
- Leverage simulation to test rare and edge-case events.
- Create modular autonomy components that can also support ADAS (driver-assist) systems.
Tech Stack & Methods
The project integrates multiple sensor modalities and learning methods:
- Sensors: Cameras, LiDAR (simulated/real), radar, GPS, and IMU.
- Models: Deep CNNs & transformers for perception, trajectory prediction networks, and RL/optimization for planning.
- Simulation: Virtual environments for training, replaying logs, and stress-testing edge cases.
- Deployment: On-vehicle compute (edge) plus cloud-based analytics for fleet learning.
Real-world Applications
- Autonomous shuttles and campus vehicles.
- Advanced driver-assistance modules: lane-keeping, adaptive cruise, collision warnings.
- Logistics and warehouse robots/vehicles for controlled settings.
- Driver behavior analytics and safety scoring tools.
Safety & Regulation
The design emphasizes redundancy, safe fallback, and compliance with emerging AV guidelines. Human oversight, logging, and explainable metrics are central to every test and deployment.
Social Plugin