V2V communication

Interactive V2V communication and cooperative driving simulation with vehicle perception, risk assessment, path clearance, multi-hop communication, and route replanning.

Language

HTML

Stars

1

Forks

0

Last push

28 Sep 2026

Problem

A vehicle's own sensors stop at line of sight. On a blind bend, a stopped car can stay invisible until the distance needed to stop has already run out.

The question the project asks: if another connected vehicle has already seen the hazard, how should that information travel, be trusted, and turn into a safe decision?

Solution

Every simulated vehicle runs its own loop of perception, fusion, risk estimation and decision-making, and exchanges observations, states and intentions over modelled V2V messages. There is no central server and no vehicle commands another; each builds its own world model and decides for itself.

Visibility is computed rather than scripted: each camera field of view is ray-cast against buildings and other vehicles, so a warning only arrives through V2V when the geometry really hides the hazard.

Architecture

1. EnvironmentRoad geometry, lanes, buildings and hazards
2. Vehicle modelsRoad-relative (Frenet) kinematics with smooth lane changes
3. Perception and V2VRay-cast camera occlusion, plus VSM, CPM, HZM, EBW, TIM and CFM messages
4. Cooperative fusionFive-gate message validation, association and existence probability
5. Prediction and riskTTC, time-to-brake, required deceleration and collision probability
6. Decision and planningPath Clearance Score, warning level, route and broadcast intent
7. VisualizationHTML5 Canvas top-down and pseudo-3D views

Features

• Blind-corner hazard awareness with geometric line-of-sight occlusion
• Six-message V2V set with geo-scoped relaying up to three hops
• Five validation gates: freshness, plausibility, signature (modelled), consistency with own sensors, and sender trust
• Path Clearance Score for keep, left, right and stop manoeuvres
• Warning levels with debounce and hysteresis; remote-only information is capped below CRITICAL until the vehicle's own sensors agree
• Intent broadcasting so following vehicles do not all take the same gap
• Route replanning after a road closure, and a degraded mode when V2V drops
• Deterministic rendering: the interactive player and the rendered video show identical numbers

Results

Reference run of the blind-corner scenario. All values are computed by the simulation, not measured on vehicles.

Limitations

• Vehicle behaviour is simulated, not measured from physical vehicles.
• Radio latency, range and message signing are modelled, not transmitted over hardware.
• The GRU/GAT prediction stage is represented by modelled outputs; no network is trained or run.
• Risk weights and thresholds are engineering choices that would need calibration with real data.
• It is an academic prototype, not a safety system.