SIH 2026ZEROIQ'S

Back to overview

PROBLEM STATEMENT 26037

Collision Intelligence

Real-Time Adaptive Navigation for Unstructured Indian Environments

PROTOTYPE DEMONSTRATION · PS 26037

Collision Intelligence Prototype - System Overview

Demonstrating the Camera + LiDAR + Radar sensor fusion and real-time obstacle detection.

Simulated Navigation on Unstructured Indian Roads

Vehicle simulation demonstrating real-time obstacle avoidance and path planning using Hybrid A* and NLMPC in chaotic traffic scenarios.

Collision Intelligence

Camera, LiDAR and Radar are fused into graded traversability, prediction, Hybrid A* and NLMPC for real-time replanning in free space.

VIEW SUBMITTED SIH PRESENTATION

EXISTING APPROACH

←→

OUR SYSTEM

drag divider
OUR SYSTEM
EXISTING APPROACH

SYSTEM VISUALIZATION

Camera + LiDAR + Radar
Perception
Sensor Fusion
Environment / Traversability
Prediction
Hybrid A*
NLMPC
Vehicle Dynamics

The problem

Path planning built for structured, well-marked roads breaks down on Indian streets, where the environment itself is the source of uncertainty.

  • Weak or absent lane markings
  • Potholes and irregular road geometry
  • Pedestrians and two-wheelers in shared space
  • Cattle and other unpredictable agents
  • Mixed, dense, and chaotic traffic behavior

Why three sensors

No single sensor is a point of failure - each covers a gap the others leave open.

CAMERA

Semantic understanding

Segmentation, agent classification, scene context

LIDAR

Geometry / traversability

Point cloud residuals, surface cost estimation

RADAR

Motion / robustness

Velocity, low-visibility resilience, redundancy

Fusion Layer

Camera + LiDAR + Radar

World Model

Real-time planning

The planner continuously re-evaluates candidate paths against a live traversability map as agents move.

SYSTEM VISUALIZATION - agents tracked this cycle

REPLANNING
VehiclePedestriansTwo-wheelersCarsCattle

Path Search

Hybrid A*

Trajectory Opt.

NLMPC