At the 2025 Horizon Technology Ecosystem Conference on December 10, Horizon's VP and Chief Architect Su Qing unveiled a paradigm shift in autonomous driving, likening Tesla's FSD V12 to the atomic bomb's proof-of-concept. Drawing from Horizon's newly mass-produced HSD (High-order Smart Driving) system, Su predicted urban L2 capabilities surging in affordability to 100,000 RMB vehicles within a year, while L4 deployment costs plummet via end-to-end data-driven models. Meanwhile, Zhixing Tech secured five parking assist projects from a top Chinese OEM, including four overseas in EU and East Asia, underscoring China's rapid globalization of smart driving tech.
Paradigm Shift: From Rules to End-to-End Learning
Su Qing highlighted how pre-2024 autonomous driving achieved only a 'half-revolution'—deep learning transformed perception, but rule-based planning kept behaviors 'unhuman-like.' Tesla's FSD V12 bridged the gap with a full end-to-end closed loop, unifying perception and control under data-driven paradigms learned from dense human driving data.
Key insights from Su:
- Product Experience Leap: Urban L2 will become 'highly human-like' in one year, dropping to 100,000 RMB ($14,000) models like automatic transmission today.
- L4 Efficiency Surge: No more high-precision maps or ODD limits; solve one complex city to cover the nation, enabling 'passenger car + Robotaxi' dual-mode at scale.
- Development 'Subtraction' Mode:剔除 bad behaviors from full human data sets, naturally emerging capabilities like autonomous roadside parking without explicit coding.
Su cautioned realism: Next 3 years focus on optimizing existing systems via compute power, larger models, and engineering—not AGI overhauls.
Horizon HSD: Mass-Producing Human-Replacement Machines
As HSD project lead, Su emphasized its goal: 'Build a machine replacing human drivers—or it's meaningless.' In 2-3 years, HSD aims for 'qualitative experience leaps' with L4-level performance at passenger car prices, democratizing advanced ADAS.
| Feature | Traditional ADAS | HSD End-to-End |
|---|---|---|
| Paradigm | Rule-based add-ons | Data-driven subtraction |
| Cost Trajectory | High (maps/ODD) | Nationwide from one city |
| Behavior | Unhuman-like | Naturally emergent |
| Timeline | Incremental | L2/L4 leap in 1-3 years |
Zhixing Tech Accelerates Global Parking ADAS Deployment
Complementing Horizon's vision, Zhixing Tech announced five production wins for its 8TOPS BEV (Bird's Eye View) parking solution from a leading Chinese brand. Four are overseas projects targeting EU and East Asia.
- Tech Stack: BEV three-in-one perception for high-res OCC (overhead obstacle) grids detecting dangling objects; Grid Map planning for dynamic paths.
- Global Edge: Munich R&D center with 25-country database (20万+ km driving, 550万+ parking images); 70+ replicated overseas scenarios.
- Adaptability: Multi-compute platform, localized for habits like head-in parking or disabled spots.
This positions Zhixing for 'drive-park-charge' integrations as smart driving hits price parity.
Why This Matters: China's ADAS Global Dominance
These announcements signal China's EV industry entering an 'experience inflection point.' Horizon's HSD targets L4 ubiquity, slashing deployment barriers amid Tesla-inspired innovations. Zhixing's wins highlight platformized, localized tech outpacing rivals—China's ADAS exports now rival its battery dominance.
Competitive landscape: XPeng, NIO, and Zeekr integrate similar stacks, but Horizon/Zhixing's scale (e.g., 8TOPS affordability) pressures Waymo/LiDAR-heavy Western firms. Globally, expect cheaper Robotaxis and L2+ in mass-market EVs, accelerating 30%+ EV adoption by 2027.
Looking Ahead: Optimization Era Dawns
Su's 3-year optimization focus aligns with industry plateauing AI theory. Horizon pushes HSD to 'seamless L4,' while Zhixing's overseas hubs fast-track ADAS exports. Watch 2025 for 100k RMB L2+ models from BYD/NIO et al., redefining 'smart' in Chinese EVs—and beyond.
