Plus’s AV2.0 Architecture for the Physical AI world
Three‑layer redundancy architecture delivers
robust physical AI for autonomous driving
robust physical AI for autonomous driving
1. Primary Driving System (Edge)
Runs on the vehicle’s primary edge computer using a REASONING-REFLEX framework to act as a virtual driver:
REASONING:
Leverages a Vision Language Model to interpret complex real‑world interactions and generate high‑level driving decisions for out-of-ODD edge cases.
REFLEX
Combines an end‑to‑end transformer for perception and motion forecast to execute rapid, real‑time maneuvers.
GUARDRAILS
Enforces safety via a transparent, rule‑based expert system that codifies human driving standards into verifiable constraints.
2. Redundant Fallback System (Edge)
Operates on a secondary edge computer that continuously monitors the primary system’s health. If a fault is detected, it triggers minimal‑risk fallback maneuvers to bring the vehicle safely to a stop.
3. Remote Operations (Cloud)
Provides mission control and human‑in‑the‑loop support for scenarios beyond the virtual driver’s Operational Design Domain (ODD), ensuring oversight and intervention when needed.
The cutting edge REASONING-REFLEX dual-model
REASONING - Strategic Driving Intelligence
REASONING uses a fine-tuned Vision Language Model for human-like understanding of complex scenes, enabling high-level decisions in rare and out-of-ODD scenarios. Its semantic reasoning enhances explainability and supports scalable deployment across diverse environments.
REFLEX - Tactical Driving Intelligence
REFLEX uses a transformer-based end-to-end model to fuse perception and motion planning for fast, real-time driving responses.
Our state-of-the-art network design allows us to provide industry-leading high-performance models that can rapidly scale up to the latest automotive grade SoCs.
Benefits of Our Approach
Interpretability
The VLM’s scene reasoning, combined with a defined perception layer, enables clearer explanations of the system’s decisions, enhancing transparency and building trust among engineers and stakeholders.
Traceability
Enhanced interpretability makes it easier to trace and rectify errors or exceptions, particularly in identifying and addressing corner cases, thereby continually improving the system.
Enhanced Safety
Driving is inherently safety-critical. Our rule-based safety envelope enables region-specific behavior and delivers strong safeguards for edge cases. Cloud-based remote operations provide human-in-the-loop oversight as a final layer of defense, ensuring ultimate safety.
Scalability
The three-layer system leverages zero-shot learning to handle out-of-sample edge cases and adapt to new environments without extensive retraining, enabling scalable deployment across diverse regions and vehicle platforms.