Yunqi Capital | Beyond the "Top Tier": The New Forces in Intelligent Driving Lighting Up the Beijing Auto Show

The Beijing International Automotive Exhibition, after a four-year hiatus, recently wrapped up. The ten-day event set multiple records: 117 global debuts, 278 new energy models, and over 1,500 exhibitors. It also saw its share of viral moments, with auto industry newcomers like Lei Jun and Zhou Hongyi drawing massive crowds.

Beneath the noise, a wave of advanced intelligent driving startups made their presence felt. As the most significant technology trend in the auto sector in recent years, intelligent driving has become a core selling point for manufacturers. From the solutions showcased by major OEMs to new products in the components zone and the dedicated Future Intelligent Driving exhibition area, there were notable examples of what comes next for autonomous driving.

Intelligent driving is one of Yunqi Capital's sustained focus areas. We believe that as autonomous driving enters a stage of intelligence-centric competition, product iteration, cost control, and scaled operations are particularly critical. At this year's Beijing Auto Show, three of Yunqi Capital's early-stage portfolio companies — DeepRoute, Renxin Technology, and Juefx Technology — showcased their latest products enabling advanced intelligent driving.

This edition of Yunqi Portfolio takes you through the intelligent driving products presented by DeepRoute, Renxin Technology, and Juefx Technology at the Beijing Auto Show.

01 Next Level: Intelligent Driving Highlights at Beijing Auto Show

More advanced and accessible intelligent driving represents a major thrust of the smart vehicle supply chain. At this year's Beijing Auto Show, numerous advanced intelligent driving technologies and smart cockpit products were unveiled, offering a glimpse of what the next level of smart vehicles will look like.

City NOA Without HD Maps

City NOA (Navigate on Autopilot) without high-definition maps was a selling point for several models. The Li Auto L6 Max, IM Motors L6, and Xiaomi SU7 Pro all adopted mapless intelligent driving solutions, claiming full-scenario NOA for both highway and urban environments.

In recent years, NOA intelligent driving solutions have become relatively mature for highway scenarios with simpler driving conditions. However, the technology has lagged in more complex urban environments. Since 2023, city NOA has gradually become the new battleground where intelligent driving vendors compete. "Mapless NOA" refers to intelligent driving that doesn't rely on high-definition maps. Given the high costs of obtaining and updating HD maps, shedding this "crutch" is a technical direction the industry is actively pursuing.

"City NOA without HD maps" means facing the dual challenge of more complex road conditions and the absence of map guidance, placing higher demands on the computing power, algorithms, and sensor precision of intelligent driving systems. In 2023, Yunqi Capital seed-round portfolio company DeepRoute launched the first intelligent driving system independent of HD maps. At this year's Beijing Auto Show, Yunqi Capital Series A portfolio company Juefx Technology also exhibited city NOA-related solutions.

"End-to-End" on the Road

Currently, common autonomous driving solutions include three steps: perception, decision-making, and execution. The "end-to-end" model simplifies these steps, directly converting data collected by sensors into vehicle execution commands. This approach is more efficient than traditional solutions but places higher demands on data training and other elements, and has not yet become widespread in the industry.

At this year's Beijing Auto Show, certain technology vendors including Yunqi Capital portfolio company DeepRoute showcased the latest results and road-test performance of their end-to-end model solutions.

Smart Cockpit

With AI large models being deployed in vehicles at scale, smart cockpit also stood out as a highlight of this year's auto show.

For instance, Xpeng Motors launched its AI Tianji system during the show, with cockpit-enabled features including an AI assistant that can automatically control air conditioning and windows, and an AI chauffeur function that can generate customized driving routes for users. SenseTime's SenseAuto exhibited an AI large-model cockpit product matrix centered on a multimodal scene brain, along with a new cockpit 3D interaction demonstration.

02 Yunqi Portfolio: Full-Stack Enablement of Advanced Intelligent Driving, from Platform to Chip

DeepRoute — Advanced Intelligent Driving Platform DeepRoute IO

During the auto show, Yunqi Capital seed-round portfolio company DeepRoute exhibited its mass-production-oriented advanced intelligent driving platform, DeepRoute IO. Independent of HD maps and applying an end-to-end model, it can recognize traffic lights, construction scenarios, and slow-moving or temporarily stopped vehicles within the coverage area of navigation maps, achieving point-to-point intelligent driving in urban areas.

DeepRoute booth at 2024 Beijing Auto Show

By the numbers, according to DeepRoute's disclosures, the IO platform achieves intersection passage success rates of nearly 98% under complex road conditions, and special intersection turning success rates of nearly 90%. The platform's perception range extends to 200 meters, with accurate identification of obstacles as small as 30cm x 30cm (roughly basketball-sized), enabling comprehensive situational awareness across all scenarios.

DeepRoute also showcased its first solution based on DeepRoute IO. The solution adopts the NVIDIA DRIVE Orin system-on-chip, with 200+ TOPS of computing power, 1 solid-state lidar, and 11 cameras, integrating driving and parking functions. Based on navigation maps, it can achieve full-domain, full-time, full-scenario intelligent pilot assistance. With this solution, DeepRoute has entered into mass production cooperation with a leading domestic OEM, with several co-developed models expected to hit the market successively in 2024.

First solution based on DeepRoute IO

IO "politely" yielding to the red vehicle behind

Renxin Technology — 16G High-Performance Automotive SerDes Chip R-LinC

Yunqi Capital Pre-A+ round portfolio company Renxin Technology launched its first 16G high-performance automotive SerDes chip product — R-LinC — during the Beijing Auto Show. This chip supports transmission rates of 16Gbps-1.6Gbps, transmission distances of up to 15 meters, and insertion loss compensation capability exceeding 30dB, with speed and process technology currently 1-2 generations ahead of peers. R-LinC has already been applied in an intelligent driving 5V super-vision solution developed in cooperation with Sony.

R-LinC achieves technological innovation in multiple areas including interface solutions. By adopting a highly integrated interface solution, it effectively simplifies network topology, reducing chip, wiring harness, and connector counts, significantly lowering system complexity and cost.

Renxin Technology booth at 2024 Beijing Auto Show

R-LinC applied in "Intelligent Driving 5V Super-Vision Solution"

Automotive SerDes chips are primarily used for vehicle video image signal transmission, playing a critical role in cameras, smart displays, and other intelligent driving and smart cockpit equipment. Transmission rate is a key performance indicator for SerDes chips — higher rates enable more complex scenario signal transmission. As competition intensifies in advanced intelligent driving technologies such as city NOA without HD maps and end-to-end models, market demand for high-speed SerDes chips is growing.

Founded in 2022, Renxin Technology has been dedicated to high-speed automotive SerDes chip R&D, and its 16G chip currently represents the industry speed ceiling. The industry faces two major pain points: technology iteration and heavy cost burdens. Renxin Technology's R-LinC pursues innovation through triple breakthroughs in "high speed, stability, and cost-effectiveness," helping the industry reduce costs and improve efficiency.

Juefx Technology — Data Closed-Loop Solution Based on Horizon Robotics Journey® 6 and BEV+Transformer Algorithm Mass-Production Architecture

During the Beijing Auto Show, Yunqi Capital Series A portfolio company Juefx Technology launched its data closed-loop solution based on Horizon Robotics Journey® 6 and BEV+Transformer algorithm mass-production architecture. This solution combines advantages in data closed-loop models, safety compliance, and cost control to establish an efficient path for breakthroughs in advanced intelligent driving mass production, helping address technical bottlenecks in the rapid development of city NOA.

The overall solution is built on Horizon Robotics Journey® 6's mass-production-oriented perception results, efficiently meeting the closed-loop prior data needs for highway NOA, city NOA, and memory commute scenarios. Vehicle-cloud collaboration breaks through local computing power and spatiotemporal limitations. The solution also provides HD automatic annotation toolchains and lightweight map closed-loop models, fully supporting OEMs in various deployment paths including "AI chauffeur," "city-by-city rollout," and "nationwide availability" through a dual-wheel drive approach.

Juefx Technology data closed-loop solution displayed at Horizon Robotics booth at Beijing Auto Show

Juefx Technology is one of the few domestic suppliers capable of providing complete mass-production vehicle data closed-loop solutions. It has successively launched a series of mass-production solutions applicable to full-scenario intelligent driving data closed-loop capabilities, including "BEV-based data closed-loop fusion intelligent driving solution" and "lightweight map highway NOA intelligent driving solution."

Technology innovation enables better living. We look forward to the day when advanced intelligent driving enters broader consumer markets worldwide and assists in daily life.