HSD V2.0 Advances End-to-End Assisted Driving

Horizon Robotics officially launched HSD V2.0 on June 29. Marking the platform’s largest over-the-air update to date, HSD V2.0 brings six major capability upgrades, 18 new features, and 25 user experience enhancements. Built on a unified end-to-end architecture, HSD V2.0 delivers point-to-point assisted driving and all-around collision mitigation, setting a new benchmark for continuous system evolution. Horizon Robotics will begin the phased rollout of HSD V2.0 across partner vehicle models. The iCAR V27 will be the first vehicle model to receive the OTA deployment starting June 30. Rollout timelines for other vehicle models will be announced separately by the respective automakers.

One Unified Model. A Leap Across Every Scenario.

Setting A New Benchmark for End-to-End Assisted Driving

HSD V2.0 goes beyond feature-level updates, introducing a major model upgrade powered by world modeling and end-to-end reinforcement learning. It delivers broad improvements in driving comfort, parking performance, all-around hazard mitigation, and personalized interaction.

Across key user scenarios, HSD V2.0 delivers comprehensive improvements. For driving comfort, the system delivers smoother, more controlled vehicle maneuvers, while overall driving and parking performance shows holistic gains. For on-road functionality, HSD V2.0 streamlines full point-to-point intelligent driving workflows, reliably handling everyday commutes and complex road environments alike. For parking, it enables more flexible parking in open and unmarked spaces with improved handling of tight, complex spots and stronger obstacle evasion. For safety, it expands active safety coverage across a broader range of objects, directions, speeds, weather conditions, and road environments. Additionally, the infotainment system now supports customizable AI-generated visual themes for the surround-view rendering interface, refreshing the entire intelligent driving display interface.

More fundamentally, as a leading end-to-end driving solution, Horizon Robotics’ HSD lineup has long focused on unlocking consistent performance across a broad range of scenarios via end-to-end modeling. HSD V2.0 leverages optimized toolchains to boost Journey® 6P computing platform efficiency, while advancing its dual-engine architecture powered by world models and end-to-end reinforcement learning. The result is a unified full-scenario driving experience defined by superior consistency, human-like driving logic, and robust cross-scenario generalization.

On top of that, HSD V2.0 delivers substantial advances in active safety via its revamped end-to-end AEB risk network model architecture. The update repurposes of the Occupancy Network (OCC) — originally built for NOA (Navigate on Autopilot) — to power active safety functions including AEB (Automatic Emergency Braking), AES (Automatic Emergency Steering), and AMAP (Accelerator Pedal Misapplication Prevention). These functions are built on a shared foundation model, and the team continuously refines training for long-tail edge cases using high-fidelity synthetic scene data to boost the model’s generalization performance.

This architecture moves active safety beyond traditional rule-based and predefined object-recognition approaches. The system can now reliably process unstructured, atypical obstacle scenarios, improving risk perception and decision-making.

Advancing the Next Generation of End-to-End Assisted Driving

While HSD V1.0 eliminated the fragmentation flaws of traditional modular intelligent driving and delivered capabilities across a broad range of driving scenarios via a unified large model — marking the first disruptive restructuring of intelligent driving architectures — HSD V2.0 builds on this foundation by further advancing its dual-engine system of world modeling and end-to-end reinforcement learning. The upgrade greatly strengthens the system’s in-depth environmental reasoning and dynamic prediction capabilities, representing a major evolution of end-to-end driving technology. It enables comprehensive leaps in the model’s understanding of physical space, dynamic behavioral logic, and risk assessment, advancing intelligent driving from reactive decision-making toward predictive understanding.

Powered by the dual-engine architecture, Horizon Robotics has established a technical foundation for continuous system improvement. The company has pioneered a continuous learning loop centered on physical world modeling. Grounded in real-world expert driving data and further enhanced by world model perception and reinforcement-learning-driven long-tail scenario optimization, the system continuously improves through data-driven training and iterative model optimization. It delivers broad capability improvements covering daily cruising, parking scenarios, general road conditions and extreme long-tail edge cases.

Unlike the fragmented, feature-by-feature updates prevalent across the industry, HSD V2.0 establishes a sustainable, self-evolving technological foundation. Supported by the world model’s robust spatiotemporal perception capabilities, it continuously narrows the gap between human driving and machine driving performance. This enables mass-produced intelligent driving systems to achieve sustained, efficient upgrades, providing a scalable path for the continued development of assisted driving.