Kognic Launches Language Grounding: Annotation Capabilities for Reasoning-Capable Autonomous Driving Models

Kognic Launches Language Grounding: Annotation Capabilities for Reasoning-Capable Autonomous Driving Models

Platform expansion addresses industry shift from perception annotation to causal reasoning, enabling next-generation end-to-end driving systems

GOTHENBURG, Sweden -- February 27, 2026 -- Kognic, the annotation platform for autonomous driving, today announced Language Grounding -- new platform capabilities that enable automotive OEMs and technology companies to produce the structured reasoning data required by next-generation driving models.

As the autonomous driving industry transitions from modular perception-prediction-planning architectures to end-to-end learning systems, annotation requirements are shifting from labeling objects in scenes to explaining why driving decisions should be made. Language Grounding addresses this shift with four new annotation modes -- Write, Edit, Rank, and Behaviour -- integrated directly into the Kognic platform.

Central to the new capabilities is Kognic's Chain of Causation (CoC) methodology, a two-step annotation workflow designed to prevent hindsight bias in reasoning data. In the first step, annotators assess driving scenes at the decision point without access to future context, ensuring observations are causally grounded. In the second step, the full sequence is unlocked for quality assurance, verifying that reasoning traces align with actual driving outcomes.

"The industry spent a decade teaching machines to see. The next decade is about teaching them to reason," said Daniel Langkilde, CEO and Co-founder of Kognic. "Language Grounding is not a new product -- it is the natural evolution of our platform, built on seven years of sensor-fusion annotation expertise. We are expanding from what is in the scene to why it matters for the driving decision."

Published research on structured causal reasoning for autonomous driving has demonstrated a 12% improvement in planning accuracy on challenging scenarios and a 35% reduction in close encounter rates in simulation, underscoring the impact of annotation quality on model performance.

Language Grounding capabilities include LLM integration for autolabel and online automation, configurable annotation interfaces per customer requirements, and support for the full annotation lifecycle from data curation through quality assurance.

Kognic serves leading automotive OEMs and technology companies developing autonomous driving and advanced driver assistance systems. The company will present its new capabilities at Tech.AD Europe in Berlin on March 23-24, 2026.

About Kognic
Kognic helps humans train machines that move. The company's annotation platform enables automotive OEMs and technology companies to produce high-quality training data for autonomous driving and ADAS systems. Founded in 2018 and headquartered in Gothenburg, Sweden, Kognic combines people, platform, and process to deliver the most annotation data for customers' budgets. Learn more at kognic.com.

Media Contact
Bjorn Ingmansson, Marketing Director [bjorn.ingmansson@kognic.com]

Lindholmspiren 7, 417 56 Göteborg, Sverige

Kognic helps humans train machines that move. The annotation platform for autonomous driving and ADAS - trusted by Bosch, Continental, and Qualcomm.

This release was published on openPR.