FLUIDCAR Fast Liquid-neUral Intelligent Driver for CARs
- Coordinator
- Ada Guzey Muhendislik Yazilim Mekatronik Sanayi Ve Ticaret Limited Sirketi, Türkiye
- Runs
- 1 August 2026 to 31 July 2028
- Status
- Running
- EU funding
- €391k
- Total cost
- —
- Partners
- 1 organisations in 1 countries
- Type
- Coordination and Support Action
- Grant agreement
- 101297156
What the project is doing
In the project's own words, from its summary on CORDIS.
FLUIDCAR develops a low-power, explainable Liquid Neural Network (LNN) that learns human steering, throttle, and braking from real driving—with no autonomous actuation. The model runs in shadow-mode: a human drives; from camera frames plus CAN/IMU context the LNN proposes control signals; we log discrepancies (e.g., steering-angle RMSE) for training and evaluation. A simple driver-ID button (1 = PI, 2 = co-driver) enables personalised models; we train two LNNs and quantify cross-driver transfer (own-model vs other-model) to measure generalisation.
Technically, we deliver a closed-form continuous-time LNN (≤2M params) optimised for edge inference <10 ms and ≤5 J per inference on Jetson Orin NX via INT8/TensorRT. Robustness is built through a three-loop process: causal saliency audits, a 300-scenario simulation stress-lab (weather, glare, sensor dropouts), and public-road shadow-mode campaigns across urban/rural, day/night, and adverse weather. Video is cached on-board, securely uploaded to HPC for training, then deleted; only models, CAN/IMU logs, and FAIR metadata are retained under GDPR-aligned procedures.
Explainability comes from neuron-level saliency and full replay of high-error events; we run an ISO 26262 orientation gap-analysis to ease later OEM integration. Outcomes include a licensable Driver-Mimic Evaluation Kit (sensor harness + edge engine + SDK), open research code and models, a cross-driver evaluation report, and an investor-ready business plan. After the project, we will sell the know-how and software to the automotive sector (Tier-1 suppliers and OEM advanced-engineering teams), with a pathway to personalised driver-twin analytics and future integration into production ADAS stacks.
Topics
Who is involved
| Organisation | City | Country | Role |
|---|---|---|---|
| Ada Guzey Muhendislik Yazilim Mekatronik Sanayi Ve Ticaret Limited Sirketi | Sivas Merkez | Türkiye | Coordinator |
Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI