World’s First Renesas V4H High-Level Domain Controller On-Board!

Recently, the first mainstream domestic model equipped with iMotion’s driving-parking integrated domain controller was officially launched, becoming one of the very few high-level smart driving domain controllers in the industry that have achieved mass production on a mid-range computing platform.

This also marks the world's first implementation of a mass-produced BEV (Bird's Eye View) + OCC (Occupancy Network) perception algorithm specifically on Renesas' V4H computing platform, establishing a new milestone for high-level autonomous driving solutions in advancing the 'democratization of smart driving'.

Powered by the Renesas V4H as its core chip, iMotion’s iDC 500 enables a range of advanced driving features, including trained commuting pilot for over 50 km, HD map-lite-based highway navigation, and enhanced cruise assistance. Additionally, the system supports memory parking for routes up to 3 km long, capable of obstacle avoidance and intelligent parking space selection along the way. It handles complex scenarios including dead-end roads and narrow spaces, completing automatic parking in under 30 seconds. and offers User-defined parking spot selection as well as a smooth HMI experience with real-time 3D modeling and rendering.

Notably, the first mass-production launch of the iDC 500 spans multiple powertrain types, including both new energy and internal combustion engine vehicles, and is aimed at a variety of domestic and international markets. The advanced parking features initially rolled out with the iDC 500 will be available across both EVs and ICE vehicles, enabling global deployment and marking another milestone in iMotion’s international expansion.

As intelligent driving continues to expand across diverse scenarios, hardware platforms capable of supporting high-level functions remain scarce. The high cost of “large computing power” has become a major bottleneck for widespread adoption and rapid iteration of intelligent driving technologies. Achieving the vision of democratized intelligent driving requires strong, synergistic efforts from both hardware platform providers and suppliers of mass-producible high-level intelligent driving solutions.

Against this backdrop, Renesas has developed the high-performance, cost-effective V4X series computing platform to meet the deployment needs of L2+ and L3 autonomous driving systems. Among them, the V4H stands out as the high-performance variant. Built on a 7nm process, the V4H offers lower power consumption and supports passive cooling, making it ideal for vehicle-side deployments where both space and energy are constrained. Its built-in R-core eliminates the need for an external MCU, enhancing on-chip efficiency while reducing power and space requirements. In addition to supporting CNN, the NPU features multiple dedicated hardware accelerators — including IMR, IMP, CVE, and DOF — as well as a programmable DSP hard core to enable flexible algorithm customization.

Nevertheless, implementing high-level functions that once required hundreds of TOPS on a mid-range computing platform delivering only tens of TOPS remains a formidable challenge. Building on the Renesas V4H, iMotion has carried out multi-layer optimization across hardware accelerators, middleware, and more to improve the overall synergy between the hardware foundation and upper-layer applications.

Leveraging the characteristics of both the model and chip resources, iMotion effectively decomposes models to maintain performance under limited resources. Custom hard-coded DSP operators are custom-coded to precisely match algorithm requirements, ensuring efficient execution. By accurately identifying critical layers prone to performance degradation and applying effective quantization techniques, we have maintained the model's performance on edge devices during fixed-point operations. Additionally, iMotion has developed a proprietary zero-copy memory scheduling mechanism that fully utilizes the chip's storage resources, meeting the latency requirements for functional safety in automotive systems.

Throughout this process, iMotion has gained deep expertise in model, chip, and operator, enabling the development of a robust upper-layer application algorithm platform. This includes a self-developed BEV Transformer + OCC perception module capable of directly outputting 3D information with centimeter-level mesh precision; the Z*2.0 parking planning algorithm, offering greater trajectory flexibility; and a prediction model based on an autoregressive diffusion network, which demonstrates strong capabilities in environmental understanding and reasoning. Combined with a spatio-temporal joint planning algorithm, these technologies work together to deliver a more comfortable, human-like driving experience.

iMotion’s extensive experience in deploying and applying advanced functions on mid-range computing platforms, along with the maturity of its upper-layer algorithm platform, enables rapid implementation of high-level intelligent driving features across diverse hardware environments. The iDC 500 series computing platforms already support a variety of chips, including the Renesas V4H and Horizon Journey 6E/M. With a dual-track strategy that integrates both domestic and international chipsets, a comprehensive product portfolio for high-level intelligent driving on mid-range computing platforms has been established — offering OEMs and consumers a wider range of choices across computing power, cost, and performance dimensions.