Adaptive Variable Resolution 2.5D Lidar Mapping for Dynamic Environment Perception
“Autonomous military vehicles need to navigate rough terrains quickly, but processing high-resolution 3D sensor data overwhelms computer processors. This project builds a smart mapping system that looks closely at what is directly in front of the vehicle for maximum safety, while simplifying data from far away to save computing power.”
Deconstructed Problem Blueprint
WHAT TO BUILD?
A high-performance deep learning pipeline and variable resolution 2.5D grid engine that transforms raw Lidar point clouds into an elevation map containing semantic layers for navigation.
WHO IS IT FOR?
DRDO, Department of Defence Production, and IDEX; used by autonomous military ground vehicles, combat support rovers, and defense engineers.
WHY DOES IT EXIST?
Full 3D Lidar data creates massive computational and memory bottlenecks in real-time execution, whereas simple 2D maps miss critical elevation data like potholes or curbs, risking vehicle safety in tactical zones.
INPUT → PROCESS → OUTPUT ARCHITECTURE PIPELINE
Raw 3D Lidar point clouds, ego-motion data from IMU/GPS, and calibration configurations.
Sparse 3D convolutions or PointNet++ for point semantic segmentation, followed by a hierarchical spatial projection algorithm to aggregate points into non-uniform 2.5D cells.
A dynamic 2.5D multi-layer elevation map with semantic color-coding (terrain, static obstacles, dynamic objects) and memory footprint analytics dashboard.
CONSTRAINTS
- Must operate in real-time (low latency / high FPS) on edge computing hardware
- Must run completely offline without relying on cloud connectivity
- Must avoid alignment errors or data loss during 3D to 2.5D multi-resolution projection
JURY METRICS
- Inference latency (processing time per frame in milliseconds)
- Classification accuracy (mIoU for terrain, static, and dynamic classes)
- Memory footprint reduction percentage compared to uniform 3D grids
- Robustness in handling sensor noise and varying environmental conditions
CORE BOTTLENECK
Designing a non-uniform spatial grid data structure that handles resolution transitions smoothly without creating projection gaps, boundary artifacts, or high memory lookup latency.
Why You Should Choose This PS (Pros)
- High strategic defense value with clear alignment to IDEX and DRDO's autonomy roadmap
- Visually stunning 'wow factor' dashboard displaying real-time 3D-to-2.5D foveated mapping
- Solves a real, mathematically rigorous bottleneck in autonomous robotics
Challenges & Risks to Watch (Cons)
- No official dataset provided, requiring reliance on open-source datasets like SemanticKITTI or nuScenes
- High technical complexity in setting up sparse 3D convolutions and custom grid data structures
- Requires efficient handling of point cloud data formats within limited hackathon timelines
Recommended Tech Stack & Grand Finale Features
Killer Features to Impress the SIH Jury
View Official SIH 2026 Portal Raw Text▼
•
Autonomous navigation depends on the ability of a vehicle to perceive its surroundings with high precision. While 3D Lidar point clouds provide rich spatial data, processing millions of points in real-time creates immense computational bottlenecks and memory latency. Conversely, standard 2D occupancy grids lose critical height information necessary for detecting curbs, potholes, or overhanging obstacles. To balance precision and performance, there is a need for a 'foveated' mapping approach—similar to human vision— where the immediate vicinity is rendered in high detail for safety, and distant areas are simplified to reduce the processing load.
•
The goal is to build a deep learning pipeline that transforms raw Lidar point clouds into a variable resolution 2.5D grid (an elevation map with semantic layers). The system must perform three primary tasks:
•
A software framework consisting of:
within a 10m radius and decreases (e.g., 50cm cells) up to a 100m radius.