Motovis

The MOTOVIS parking-driving integrated domain controller is a high-performance integrated platform fully developed in-house by MOTOVIS for mid-to-high-level intelligent driving needs. It covers mainstream highway L2+ driving functions and supports highway and urban NOA as well as advanced memory parking (HPA), improving scenario coverage while enhancing system safety. The new-generation MagicPilot2.0 adopts an ultra-lightweight BEV parking-driving integrated network and software design, delivering a high-performance, all-time driving and parking experience; combined with a high-performance, low-power domain controller, it can be standard equipment even on entry-level models and supports overseas deployment, making it the preferred choice for vehicle export programs.

MagicPilot2.0: ultra-lightweight BEV parking-driving integrated network and software design; 5V12U optimal sensor combination (5 cameras + 12 ultrasonic sensors); supports high-performance, low-power domain controller form factor; supports overseas deployment. Magic Pilot: single SoC integrates driving and parking functions, supporting 6 cameras, 5 millimeter-wave radars and 12 ultrasonic sensors, with a proprietary Pre-Fusion framework for multi-sensor data pre-fusion; developed to ISO 26262 ASIL-D functional safety; supports mainstream SoC and MCU platforms with expansion to higher-level differentiated autonomous driving functions. Driving functions: full-speed adaptive cruise control (ACC), automatic emergency braking (AEB), lane keeping assist (LKA), lane centering control (LCC), traffic jam assist (TJA), integrated cruise assist (ICA), digital video recorder (DVR), highway navigate on pilot (NOA HW). Parking functions: 360 surround view and sentry mode (SM), parking distance control (PDC), fully automatic fused parking assist (FAPA), remote parking assist (RPA), reverse driving assist (RDA), home parking assist (HPA).

• Highway L2+ driving scenarios for passenger cars
• Highway and urban NOA navigation assist driving
• Fully automatic fused parking and remote parking
• Advanced memory parking (HPA)
• Standard intelligent driving for entry-level to mid-high-end passenger cars
• Export models with overseas deployment of driving and parking functions

The MOTOVIS fused parking domain controller MagicParking3.0 adopts a new parking 3.0 algorithm platform, using spatio-temporal fused BEV perception and highly dynamic joint planning and control to deliver a human-like, smooth parking experience. Developed on high-quality parking scenario data from more than 60 regions worldwide, its global deployment has been fully validated, and it supports up to 11 highlight functions, forming a complete parking-domain solution suite.

New parking 3.0 algorithm platform; spatio-temporal fused BEV perception; highly dynamic joint planning and control; supported by high-quality parking scenario data from more than 60 regions worldwide with fully validated global deployment; supports up to 11 highlight functions (the official page does not itemize all function names or specific compute/pixel figures).

• Automatic parking assist (APA) for passenger cars
• Diverse parking scenarios including complex and extremely narrow spaces and mechanical parking slots
• Deployment and validation of parking functions for global vehicle models
• Human-like parking experience with highly dynamic planning and control

The MOTOVIS driving domain controller and front-view integrated camera product family is represented by the MagicDrive2.0 L2 intelligent driving integrated camera, which adopts a new-generation 4D BEV perception network to achieve accurate and continuous environment perception and prediction, making driving smooth and stable. The product meets C/ENCAP, functional safety and strict domestic and international regulatory requirements, safeguarding the bottom line of driving safety, and supports a single-camera (1V) product form for building standard ADAS products for vehicle models.

New-generation 4D BEV perception network (driving domain); supports single-camera (1V) product form; meets C/ENCAP requirements, functional safety requirements and strict domestic and international regulations; corresponds to L2 intelligent driving functions (the official product page does not publish specific TOPS compute, megapixel or interface figures, so no specific numbers are stated).

• L2 intelligent driving assistance for passenger cars
• Standard ADAS (standard-equipment) solutions for vehicle models
• Export models meeting C/ENCAP and domestic and international regulations
• Volume production of OEM L2 intelligent driving functions

The MOTOVIS commercial vehicle intelligent driving product series addresses the industry characteristics of commercial vehicles (medium and heavy trucks, light trucks, mini trucks) with high accident rates, severe consequences and safety as a core necessity. The commercial vehicle AEB series within Magic Drive delivers highly reliable solutions tailored to the large body, numerous blind zones and long braking distances of commercial vehicles. The products have passed strict commercial vehicle regulatory certification and have been delivered in volume to several leading domestic commercial vehicle OEMs, exploring the commercial vehicle market together with partners such as Shaanxi Heavy Duty Truck, Foton, Beiben and FAW. The main front-mounted products are the monocular front-view integrated camera and radar fusion system, supporting dual warning, AEB, LKA and other regulatory functions, and can provide perception and localization solutions required for L3/L4.

For commercial vehicles such as medium/heavy trucks, light trucks and mini trucks; monocular front-view integrated camera and radar fusion system form factor; supports dual warning, AEB, LKA and other regulatory functions; can provide perception and localization solutions for L3/L4; has passed strict commercial vehicle regulatory certification (the official page does not publish specific compute, pixel or interface figures).

• Front-mounted intelligent driving and safety configuration for medium/heavy trucks, light trucks and mini trucks
• Regulation-compliant mass production of mandatory commercial vehicle AEB
• Implementation of regulatory functions such as dual warning, AEB and LKA
• Perception and localization solutions for L3/L4 advanced autonomous driving
• Front- and rear-mounted active safety applications for commercial vehicles

The MOTOVIS MoDun commercial vehicle monitoring terminal is an automotive-grade full-stack AI + connected solution for commercial vehicles, supported by advanced artificial intelligence algorithms and integrating blind spot detection (BSD), driver behavior analysis (DMS) and front-view warning (ADAS). Targeting the safety pain points of urban transport sectors such as buses, muck trucks and hazardous chemical tankers, it shifts from post-accident emergency response to proactive pre-accident warning and prevention for drivers. The MoDun blind spot active safety system recognizes pedestrians and non-motorized vehicles in vehicle blind zones and triggers automatic alarms as soon as they enter the warning area, promptly alerting the driver. The MoDun full-stack AI algorithm software meets customers' one-stop needs for long-term development and iterative evolution of computing chip platforms, powering the new era of AI + intelligent connected integrated terminals. The MoDun full-stack automotive-grade front-view ADAS integrated camera fully complies with the JT/T1178 and JT/T1242 standards of China's Ministry of Transport, providing a complete one-stop fusion solution.

Full blind zone coverage: 100-degree wide-angle cameras on the right, left and front sides fully covering the vehicle's blind zone range; high recognition accuracy: comprehensive prediction and precise warning based on multiple dimensions such as vehicle speed, distance and TTC, supporting AEB and collision mitigation; strong scenario adaptability: suitable for night and mixed pedestrian-vehicle scenarios, accurately identifying drivable space and effectively reducing false alarms; algorithm software based on deep learning, flexibly configurable on demand, supporting various advanced computing chips and system integration; front-view ADAS integrated camera complies with JT/T1178 and JT/T1242 standards, supporting international and domestic 77G radar / ultrasonic radar fusion.

• Urban transport vehicles such as buses, muck trucks and hazardous chemical tankers
• Active safety warning for pedestrians and non-motorized vehicles in commercial vehicle blind zones
• Driver behavior analysis (DMS) and fatigue/distraction monitoring
• Front-view warning (ADAS) with AEB and collision mitigation
• Front- and rear-mounted applications for heavy trucks, light trucks, coaches and buses
• In-vehicle AI edge computing and AI + intelligent connected integrated terminals

The MOTOVIS full-stack autonomous driving solution includes complete systems as well as a full core algorithm suite deeply optimized for computing chips, forming three core product lines covering passenger car front-mounting, commercial vehicle front-mounting, and the aftermarket active safety market. The passenger car front-mounted product line mainly covers various intelligent parking and intelligent driving products for passenger cars, covering L1 to L4 autonomous driving functions, with customers including more than 90% of mainstream passenger and commercial vehicle OEMs in China.

Product forms include complete systems and a full core algorithm suite deeply optimized for computing chips; passenger car front-mounting covers L1 to L4 autonomous driving functions; corresponding to intelligent parking and intelligent driving functions (the official page does not publish specific compute, pixel or interface figures).

• Front-mounted intelligent parking functions for passenger cars
• Front-mounted intelligent driving functions for passenger cars
• Autonomous driving function configurations covering L1 to L4
• Platform-based continuous iteration and mass production for vehicle OEMs

The Magic Wheel new-generation data platform is the data foundation among MOTOVIS's three major technical foundations. The official website describes it as a new-generation data platform featuring 'data-driven, rapid iteration', connecting global fleets 24/7, realizing automated data processing through cloud-based large models, with algorithms iterating through continuous feedback from the real physical world. It is the mass-production data carrier in MOTOVIS's 'mass-production data + world model' dual-drive strategy.

Full-domain connectivity: 24/7 interconnection with shadow mode automatically capturing edge scenarios; intelligent processing: cloud-based large models realize automated data processing, improving efficiency by up to a hundredfold; rapid iteration: from data feedback to model version iteration in as fast as 7 days.

• Mass-production data feedback and data closed loop for global production fleets
• Automatic capture of corner cases in shadow mode
• Cloud large-model-driven automated data annotation and processing
• Rapid model version iteration and mass-production validation feedback

The World Model is one of MOTOVIS's three frontier AI achievements, officially positioned as 'the paradigm for reconstructing one-stage end-to-end architectures' and belonging to the large-model foundation among MOTOVIS's three major technical foundations. The official website states that it is deeply integrated with the world model and VLA architecture, enabling the system to understand physical causality and common sense, moving from 'seeing' to 'foreseeing', becoming an intelligent agent that drives physical entities, with video scene transfer and behavioral data generation capabilities.

Video scene transfer: capable of transferring original video scenes across different weather conditions (sunny, rainy, snowy, foggy, dusty) and lighting conditions (daytime, night), demonstrating the system's perception and decision stability in extreme environments; behavioral data generation: capable of generating different traffic behavior data and performing operations such as lane changes, demonstrating the system's adaptability to diverse traffic rules and driving habits; deeply integrated with the VLA architecture for understanding physical causality and common sense and inferring future changes of physical scenarios.

• Validation of perception and decision stability under extreme weather and nighttime lighting conditions
• Data generation and simulation of diverse traffic behaviors such as lane changes
• Paradigm reconstruction of one-stage end-to-end autonomous driving architecture
• Prediction of motion patterns of surrounding objects and inference of future changes in physical scenarios
• Embodied AI platform construction in synergy with the data platform and reinforcement learning

The latest-generation end-to-end autonomous driving architecture is one of MOTOVIS's three frontier AI achievements. The official website states that 'even pure vision can predict the future', with an implicit world model at its core creating an in-vehicle 'digital brain' that completes detection, tracking, mapping, prediction and planning synchronously through a single model, realizing a compute-efficient, minimalist end-to-end intelligent driving solution validated as stable and reliable on real vehicles.

Lightweight input: perception, prediction and planning achievable with cameras alone, with flexible LiDAR augmentation for complex urban road conditions; anticipatory decision-making: learns the motion laws of the physical world to handle sudden situations such as ghost pedestrians and cut-ins, anticipating in advance rather than remedying afterwards; five-in-one model: a single model synchronously handles detection, tracking, mapping, prediction and planning, compute-efficient with a minimalist architecture, validated stable and reliable on real vehicles; core is an implicit world model. (The official website does not publish the architecture's specific compute, parameter counts or sensor figures.)

• Pure-vision end-to-end intelligent driving in complex urban road conditions
• Anticipatory decision-making for sudden situations such as ghost pedestrians and cut-ins
• High-level perception configuration enhanced by LiDAR
• In-vehicle deployment integrating detection, tracking, mapping, prediction and planning
• Mass-production implementation and real-vehicle validation of end-to-end intelligent driving solutions

VLA (Vision-Language-Action model) is a core component of MOTOVIS's large-model technology foundation. The official website states that it fuses vision, language and action to understand complex traffic like an experienced driver. From the early CNN to today's VLA and world model, the MOTOVIS team has invested continuously, using VLA to enable the system to understand physical causality and common sense, allowing intelligent driving to truly 'comprehend' the real world.

Fuses the three modalities of vision, language and action; deeply integrated with the world model to understand physical causality and common sense; human-like understanding and decision-making for complex traffic scenarios; technical route evolving from CNN to VLA and world model (the official website does not publish model parameter counts, training data scale or other figures).

• Human-like driving understanding and decision-making in complex traffic scenarios
• Multimodal intelligent driving fusing vision, language and action
• Physical causality and common-sense reasoning in synergy with the world model
• General algorithmic capability foundation for physical AI

The MOTOVIS full-stack proprietary algorithm platform is MOTOVIS's 'algorithm' technology foundation. The official website states that it is fully self-developed from perception, fusion and localization to planning and control, with no technical weak points. The four core algorithm capabilities have been validated on millions of real-world scenarios, forming core algorithm assets for physical AI and making MOTOVIS one of the few leading technology companies in the automotive industry that independently owns full-stack algorithms from perception to planning and control. The platform covers the core full-stack capabilities of algorithm model design, compute deployment optimization and data closed loop, and can independently empower a wide range of industries.

Four core algorithm capabilities: 1) 4D spatial perception - real-time perception of multi-sensor heterogeneous data, accurately understanding the spatial logic of the real physical world; 2) world model and dynamic prediction - able to predict the motion patterns of surrounding objects and infer future changes of physical scenarios; 3) automotive-grade highly reliable edge deployment - realizing efficient, low-latency, high-safety implementation of intelligent algorithms on physical hardware; 4) complete data closed-loop iteration system - enabling continuous self-evolution of algorithms in real scenarios while greatly reducing the cost of data acquisition. Core full-stack capabilities: algorithm model design, compute deployment optimization, data closed loop.

• Mass-production algorithm delivery for passenger and commercial vehicle intelligent driving
• Real-time fused perception of multi-sensor heterogeneous data
• Automotive-grade highly reliable edge algorithm deployment
• Continuous algorithm iteration driven by mass-production data closed loops
• Algorithm empowerment for embodied AI and other physical AI scenarios

The MoDun full-stack AI algorithm software is an automotive-grade algorithm software product launched by MOTOVIS for commercial vehicles that can be licensed separately (alongside the MoDun blind spot active safety system and the MoZhi front-view ADAS integrated camera, it is one of the three product categories of the MoDun full-stack AI + connected solution). The software meets customers' one-stop needs for long-term development and iterative evolution of computing chip platforms, thoroughly overturning the traditional bulky in-vehicle terminals based on discrete devices or core boards, powering the new era of AI + intelligent connected integrated terminals. The algorithms are based on deep learning and can be flexibly configured as needed, supporting various advanced computing chips and system integration.

Delivered as software and licensable separately; meets customers' one-stop needs for long-term development and iterative evolution of computing chip platforms; based on deep learning; functions flexibly configurable on demand; supports various advanced computing chips and system integration; replaces traditional bulky in-vehicle terminals based on discrete devices or core boards; functions cover three application categories: blind spot detection (BSD), driver behavior analysis (DMS) and front-view warning (ADAS) (the official website does not publish compute and accuracy figures of the algorithm models).

• Algorithm software licensing and integration for commercial vehicle AI + intelligent connected integrated terminals
• Long-term development and iterative evolution of OEM/Tier1 computing chip platforms
• Software deployment of blind spot detection (BSD) algorithms
• Software deployment of driver behavior analysis (DMS) algorithms
• Software deployment of front-view warning (ADAS) algorithms
• Upgrading and replacing traditional in-vehicle terminals based on discrete devices or core boards

The Embodied AI and physical AI agent platform is MOTOVIS's agent-platform direction extending from intelligent driving toward broader physical AI scenarios. The official website states that MOTOVIS, through its three capabilities of environmental perception, behavior prediction and planning control, enables the automobile to evolve from a 'mechanical tool' into an 'intelligent agent that understands the physical world'; and based on deep collaboration on the data platform, reinforcement learning and the world model, it builds an embodied AI platform together with partners, enabling other intelligent agents to autonomously navigate complex terrain and dynamically avoid obstacles, becoming all-round partners for humanity. The company's official slogan is From Embedded AI To Embodied AI (embedding AI in the real world and letting intelligence drive physical entities).

Three capabilities: environmental perception - 4D BEV driving and parking perception algorithms that analyze dynamic elements in space in real time and accurately construct a 'live map' of the physical world, letting the car 'see clearly, understand and see far'; behavior prediction - learning based on physical laws to anticipate pedestrian and vehicle dynamics, achieving 'one step ahead' proactive decision-making; planning control - lateral and longitudinal joint control that brings driving strategy closer to 'human wisdom', balancing efficiency and safety. The platform is built on deep collaboration among the data platform, reinforcement learning and the world model, enabling other agents to autonomously navigate complex terrain and dynamically avoid obstacles. The official website does not publish specific hardware forms or figures.

• Embodied AI platform construction and joint development with partners
• Autonomous navigation of complex terrain and dynamic obstacle avoidance by physical AI agents
• Agent applications in industrial production, urban operation and daily-life scenarios
• Perception-decision-action closed loop of the automobile as an embodied AI physical entity
• Physical AI capability expansion driven by 'mass-production data + world model'