From Raw Signals to Smart Decisions

A BMW Group research initiative is leveraging the COVESA Central Data Service Playground and contributing to the Data Architecture / Infrastructure Pillar to explore how vehicle data can drive smarter, real-time decisions through explainable, rule-based reasoning.

Together with Renesas and RemotiveLabs, whose RemotiveCloud platform acts as a VSS-based data feeder, the project is building the infrastructure needed to realize the DIKW (Data, Information, Knowledge, Wisdom) model. It defines standardized interfaces for cross-platform data management and establishes an open framework for semantic reasoning.

COVESA community participation is encouraged to help shape this shared and transparent approach to decision-making in software-defined vehicles.

From Data to Knowledge: What the Project Aims to Do

The team is moving beyond basic signal translation (like CAN to VSS). Instead, they’re designing a layered architecture that enables real-time inference based on structured, rule-driven logic.

“We want to act on vehicle data in real time to build rules directly into the knowledge layer, enabling transparent, human-readable decisions,” said Christian Mühlbauer, BMW Research Data Architecture Team. 

Rather than relying solely on machine learning, the team applies Datalog and a symbolic reasoner (RDFox, a high-performance reasoning engine used for logic-based inference) to encode logic, for example, identifying aggressive driving based on combinations of steering angle, speed, and time.

“This approach makes reasoning more explainable, flexible, and domain-independent,” added Haonan Qiu, BMW Research Data Architecture Team.

The data architecture follows the DIKW (Data → Information → Knowledge → Wisdom) model, with each layer adding structured logic and inference capabilities on top of the raw signals provided.

Two example driving segments used in the showcase feature average steering changes of 225° and 249.5°, reinforcing the validity of the rule logic.

🟩[Add:] Rules can also be updated at runtime, allowing dynamic experimentation and adjustment of thresholds within the Playground environment.

To explore the structure and logic behind this in more detail, check out the Knowledge Layer Hello World example on GitHub.

The Role of the COVESA Central Data Service Playground (CDSP)

The COVESA Central Data Service Playground (CDSP) serves as the collaborative backbone of the project. It provides a neutral, open environment where researchers can prototype and refine ideas using real or simulated data.

Until now, CDSP has focused on the Data and Information layers of the DIKW (Data → Information → Knowledge → Wisdom) model, structuring and contextualizing raw vehicle signals. With the BMW initiative, the project is now extending CDSP to include the Knowledge and Wisdom layers, allowing real-time reasoning and inference logic to be applied directly to the data flow.

What the Project Contributes

• Shared environment for time-series data ingestion and storage
  Foundation for the Data layer, enabling structured access to raw signals from vehicles.
• Set of standardized interfaces for the Information layer
  Enables interoperability across components, platforms, and teams by structuring raw data into meaningful, context-aware formats (e.g., VSS).
• Tools and logic engines for building the Knowledge layer
  Contributing reasoning capabilities (e.g., Datalog + RDFox) that apply rule-based logic to detect and interpret complex driving behaviors and other conditions.
• Open framework for decision logic in the Wisdom layer
  Working toward a modular, transparent system where real-time decisions can be explained, tested, and trusted, empowering software-defined vehicles.

“What’s powerful about the Playground is that it allows teams to plug into a shared ecosystem and experiment openly, without building everything from scratch,” said Stephen Lawrence, Principal Engineer at Renesas.

From raw data to wisdom, the layered architecture of semantic reasoning in CDSP shows how signals from RemotiveLabs feed into increasingly intelligent representations via RDFox and Datalog.

Combining the CDSP with RemotiveLabs Solution

While the project focuses on advancing the knowledge and inference layers, RemotiveCloud, available via its free tier, plays a key enabling role by translating raw vehicle signals into VSS and populating the Information layer of the Playground. This structured, cloud-accessible data serves as the foundation for real-time inference and semantic reasoning.

RemotiveCloud feeds vehicle signal data into the COVESA Central Data Services Playground (CDSP). The system processes the data through a rule-based engine and outputs an inference, highlighting an example like aggressive driving detection.

RemotiveCloud enables:

• Easy signal subscription and conversion across different VSS versions
• Scenario-based input configuration for reasoner benchmarking
• A low-friction way to kickstart experimentation in semantic reasoning

“When data is concrete and ready to use, it’s a huge motivator to start a research project,” said Haonan Qiu, BMW Research Data Architecture Team. “The approach with vehicle recordings in the cloud lowers the barrier for experimentation.”

“With more contributors, we can benchmark richer use cases and grow the community,” said Christian Mühlbauer, BMW Research Data Architecture Team.

The RemotiveLabs feeder integration is available as an open-source example on GitHub.

Expanding the Horizon: From Rule-Based Inference to Neuro-Symbolic AI

This initiative is just the start. The team is expanding CDSP’s capabilities with:

• New signals and domain models
• Diverse driving simulations
• Robustness testing for inference engines
• Behavioral modeling (e.g., passenger scenarios)

Next, the goal is to blend symbolic logic with statistical approaches and LLMs, moving toward Neuro-Symbolic AI, a hybrid framework combining deep learning and knowledge-based reasoning for powerful yet explainable decision-making.

“Neuro-Symbolic AI is a burgeoning field that marries two distinct realms of artificial intelligence: neural networks, which form the core of deep learning, and symbolic AI, which encompasses logic-based and knowledge-based systems.”
 — AllegroGraph

Future development also includes improving VSS modeling for multi-domain reasoning and exploring alternative open-source reasoners to supplement RDFox.

How to Get Involved and Contribute

Visit the project home, join the project weekly meetings on Wednesdays at 8 am PT, and contribute in the following ways:

• Share vehicle recordings, solely in VSS format, required to enrich the data available in the Playground.
• Collaborate on domain models and rule sets, bringing your use cases and reasoning needs.
• Test and integrate components to validate interoperability and real-time inference capabilities.

A “COVESA Edition” of RemotiveCloud has also been proposed. If OEMs commit to sharing anonymized vehicle data, this edition would offer richer datasets than the current free tier, fueling deeper collaboration across the ecosystem, from OEMs and Tier 1s to developers, UX designers, and academic researchers.