OEMs and Suppliers' Embodied Artificial Intelligence (and AI Robot) Layout Trend Report, 2024-2025

From Automobiles to Embodied Artificial Intelligence (EAI): Differentiated Layout Strategy amid Industrial Correlation

1. EAI Layout: Multiple Paths of OEMs and Suppliers

In the wave of Embodied Artificial Intelligence (EAI), OEMs and suppliers in the automotive industry chain have set foot in the industry and vigorously explored development strategies that suit them. At present, they deploy EAI by way of independent R&D, cooperative R&D, investment and cooperative application exploration. Companies need to choose their specific layout models, each of which has its own advantages and disadvantages, according to their own resources, technical strength and market goals.

Simply put, EAI is an agent that perceives and acts based on the physical body. It is like giving a robot a 'body' that can interact with the environment, so that it can learn and perform tasks by observing, moving, and speaking like humans. The concept has brought unprecedented opportunities and challenges to the automotive industry.

(1) With coherent technological development, OEMs are accelerating the deployment of humanoid robots:

OEMs' EAI deployment focus on technology reuse, supply chain collaboration, and market incremental mining. Moreover, intelligent driving technology (such as perception algorithms and decision-making models) is highly homologous to EAI. OEMs can draw the AI capabilities they have accumulated in the field of autonomous driving on robot development to reduce R&D costs. In the future, as hardware costs decline and foundation model capabilities improve, EAI will gradually penetrate from industrial scenarios into the consumer sector, becoming a key pillar for the intelligent transformation of OEMs.

On November 6, 2024, Xpeng unveiled the next-generation humanoid robot 'Iron' at the 'XPENG AI Day'. It has entered the Guangzhou factory to participate in the production and training of Xpeng P7+. In the future, it will focus on scenarios such as factories and offline stores.

Iron is designed with a humanoid structure, with a height of 178cm, a weight of 70kg, and a 1:1 human hand size design. It has 15 degrees of freedom and supports tactile feedback; it has more than 60 joints and can simulate a variety of human actions, such as standing, lying, sitting, etc.

Equipped with Xpeng AI Eagle Eye Vision System, Iron boasts 720° environmental perception capabilities without blind spots, and also uses end-to-end foundation models and reinforcement learning algorithms.

Fitted with the Xpeng AIOS which is also seen in Xpeng’s vehicles, Iron can communicate fluently and freely, has memory and can reason.

It uses a Turing AI chip, which is specially customized for foundation models. It has a 40-core processor with a computing power of 3000T. This chip is the first chip that can be used in AI cars, AI robots and flying cars simultaneously.

(2) Suppliers are making every effort to deeply lay out the EAI industry chain

Automotive suppliers rely on existing hardware technologies (such as sensors, chips, motors) and supply chain resources to extend into the field of EAI. Their core logic lies in technology reuse and collaborative cost reduction. For example, the LiDAR vendor RoboSense has adapted its automotive perception solution to robots so as to migrate its environment modeling capabilities; motor companies have reused automotive powertrain technology to develop high-density joint drive modules. This move can leverage mature manufacturing experience and customer network, but it has to address the stringent requirements of robotic scenarios for hardware flexibility.

On January 3, 2025, RoboSense hosted its first AI Robotics Global Online Launch Event of 2025, titled “Hello Robot”, comprehensively presenting RoboSense's robotics technology platform strategy, and releasing a variety of digital LiDAR sensors, robot vision products and incremental parts and solutions for robots.

At the press conference, RoboSense launched the second-generation Papert 2.0 dexterous hand which incorporates 20 degrees of freedom, a maximum load of 5 kilograms and 14 force sensors on the fingertips and palms. Together with the robotic arms and control system, it can flexibly replicate the fine movements and operations of the human hands, such as delicately picking up eggs and screws.

In addition, RoboSense demonstrated for the first time its self-developed humanoid robot, which is defined as a universal parts development platform for R&D of incremental components and solutions of robots. Based on complete machines, RoboSense will focus on three types of incremental parts including vision, touch, and joints of robots to empower the robot industry. RoboSense also announced products such as the FS-3D force sensor for end-point motion control of legged robots, the LA-8000 high-power density linear motor, and the DC-G1 highly integrated, small-sized, high-computing-power, low-power robot domain controller.

(3) Technical talents in the field of autonomous driving dabble in EAI

Similar technical paths and industry dividends stimulate autonomous driving practitioners to dabble in EAI. After L4 autonomous driving encountered obstacles in commercialization, capital and talent shifted to EAI. Autonomous driving and EAI highly overlap in perception algorithms and decision models (end-to-end reinforcement learning), so that they can share algorithms. Autonomous driving practitioners excel at AI algorithms and quick iterations, but they need to learn knowledge about hardware interaction such as mechanical control (such as force feedback and motion planning), and face the challenges like uncertainty in technical routes and commercial verification. Their aims to dominate early EAI technology through collaborative innovation of software and hardware.

In November 2023, WeRide's former COO Zhang Li announced that he would join LimX Dynamics as chief operating officer (COO). Zhang Li has extensive experience in the field of autonomous driving and has led WeRide to perform remarkably in the commercialization and technology R&D of robotaxis. His joining marks LimX Dynamics’ further efforts in the field of EAI, and also reflects the in-depth integration trend of autonomous driving and EAI.

In October 2024, LimX Dynamics launched TRON 1, a 'three-in-one' modular polymorphic bipedal robot designed specifically for mobility and manipulation in complex environments. Advanced motion control algorithms and high-precision sensor fusion technology enable TRON1 to walk stably in unstructured terrain (such as stairs and gravel roads) and perform difficult actions such as climbing slopes and overcoming obstacles. The modular design allows users to flexibly configure functional modules according to task requirements, such as installing robotic arms for grasping or integrating visual sensors for environmental perception and navigation. The core advantage of TRON1 lies in its highly flexible movement and strong environmental adaptability, which gives it wide application potential in industrial inspection, emergency rescue, logistics and distribution and other fields.

2. Robotics and automotive industries: collaborative development with similarities and differences

(1) Industrial commonality: collaborative foundation of technology and supply chain

Hardware: The robotics and automotive industries involve highly similar hardware components. Motors, sensors, deceleration/conversion mechanisms, batteries, bearings, structural parts, cooling systems, controllers, chips and other hardware are widely used in both industries. For example, the design of Tesla Optimus borrows heavily from automotive hardware technology.

Tesla Optimus has a total of 14 rotary actuators, each of which uses 2 angular contact ball bearings and a cross roller bearing; each of its 14 linear actuators has a four-point contact bearing and a ball bearing. These four types of bearings have been widely used in the automotive industry.

Sensors: LiDAR, cameras, radar, etc. all play key roles in both autonomous driving and robotic navigation. Vehicles use LiDAR to accurately detect obstacles ahead and recognize road boundaries to achieve autonomous driving. AI robots scan the surrounding environment by LiDAR to flexibly avoid obstacles and navigate accurately. In addition, automotive ECUs and robot motion controllers share the same underlying logic. They both receive sensor signals, quickly calculate according to the preset algorithms, and issue instructions to actuators.

The Xpeng AI Eagle Eye Vision System mounted on Xpeng Iron relies on Xpeng's intelligent driving system, which combines end-to-end foundation models and reinforcement learning algorithms. This vision system allows Iron to observe the surrounding environment within a 720° range without blind spots.

Software technology: The algorithms accumulated by OEMs in the field of autonomous driving provide valuable experience for the development of EAI. In autonomous task processing, humanoid robots and autonomous vehicles follow the process of 'perception - decision-making - execution', and they are the same to a certain extent at the model level. Key algorithms for path planning and motion trajectory prediction, intelligent driving algorithms can be reused on humanoid robots.

For example, Tesla applies the vision, navigation and AI algorithms in FSD to Optimus, which can thereby perceive the environment and make autonomous decisions. Benefiting from Tesla's increasingly powerful AI training capabilities and autonomous driving scenario simulation systems such as Dojo, Tesla’s robots will have the ability to recognize environmental paths and surrounding objects and plan paths before leaving the factory.

Supply chain: The mature experience of the automotive supply chain provides strong support for the development of the robotics industry. The robotics industry chain and the automotive supply chain have the same technical origins in some parts and components, such as batteries, motors, bearings and so on. After long-term development, the automotive supply chain has experience in large-scale automated production and can help achieve mass production of robots with lower costs.

(2) Industrial difference: differences in market demand and technical focus

Supply chain - custom parts:

Automobiles: Thanks to highly standardized automotive production, many auto parts are universal and can be mass-produced to reduce costs.

Robots: For diverse application scenarios, EAI products vary in forms and functions, so that their parts are mostly customized. For example, robots used for medical surgeries and household cleaning robots whose components such as joint structures and sensors vary greatly due to different functional requirements are difficult to see large-scale standardized production.

Supply chain - response speed:

Automobiles: The automotive production cycle is long, and it may take several years from design to mass production. The relatively stable supply chain does not need respond in a very short time. OEMs usually make production plans in advance and sign long-term contracts with suppliers to ensure stable supply of parts.

Robots: Rapid changes in market demand and frequent technological updates and iterations require the supply chain to respond faster. For example, consumer-grade smart robot manufacturers have to quickly adjust product designs once new functional requirements or technological breakthroughs emerge in the market.

Hardware technology - integration and complexity:

Automobiles: An automobile means a highly integrated and complex system with many hardware components working closely together. Engines/motor systems, electric drive systems, battery systems, transmissions, chassis, electrical systems, etc. all involve complex designs and functions, and they should adapt to each other.

Robots: Although multi-hardware collaboration is also involved, the integration method and focus are different from automobiles. EAI products may focus more on the integration of specific functional modules, such as robot joint modules integrating motors, sensors and controllers.

Hardware technology - energy storage:

Automobiles: New energy vehicles use power batteries as their main energy storage devices, such as lithium-ion batteries. They need huge battery capacity which is usually measured in kWh for the purpose of a long cruising range, usually around a few hundred kilometers.

Robots: Diverse and flexible energy storage hardware can be selected according to product functions and application scenarios. Compared with new energy vehicles, the energy storage hardware of EAI products should feature miniaturization, lightweight and adaptability to different scenarios, rather than a long cruising range.

Software technology - algorithms and data processing:

Automobiles: In the field of autonomous driving, algorithms center on environmental perception, decision-making and planning. For example, object detection and recognition algorithms recognize vehicles and pedestrians on roads, and path planning algorithms plan safe driving paths based on sensor data. Data processing is mainly about massive but stereotypical road information collected by automotive sensors.

Robots: Algorithms cover a wider range of fields. In addition to environmental perception and motion planning, they also include interactive algorithms for natural language processing and emotion recognition. Data comes from various sources, including not only sensor data, but also user interaction data, cloud data, etc.

3. Layout: key considerations for OEMs and suppliers

Technology R&D and innovation:

They should invest heavily in core technologies of EAI, such as foundation models, sensor fusion, motion control, human-computer interaction and other fields. For example, in the research and development of Optimus, Tesla has continuously optimized the effects of migrating FSD technology to enhance the robot's perception and decision-making capabilities; Huawei continues to iterate the Pangu Models to enhance task planning and multi-scenario generalization capabilities for EAI.

OEMs and suppliers should avoid duplicating research and development, as the automotive and robotics industries have significant commonalities in many aspects. In terms of supply chain, the raw materials and parts supply channels required by the two often overlap. Integrating supply chain resources can significantly reduce procurement costs and management difficulties. At the software technology level, AI foundation models and deep learning algorithms are the core driving forces for the realization of EAI. Sharing these technical achievements can accelerate the research and development process. As for hardware technology, sensors perceiving the environment as well as motors, gears, bearings, etc. which handle power transmission and mechanical movement are highly similar. Unifying technical standards and R&D plans can avoid repeated R&D investment.

Market demand and application scenario expansion:

In the emerging EAI field, market demand and application scenarios are still being explored and expanded. Enterprises should delve in market surveys, gain an in-depth understanding of the demand of different industries and users, and develop targeted products and solutions. They should actively cooperate with potential customers, carry out pilot projects and application demonstrations, accumulate market experience, and improve the market adaptability and competitiveness of their products.

In addition to common scenarios such as industrial manufacturing, logistics warehousing, and home services, they should vigorously explore the application potential of EAI in sectors such as medical care, education and agriculture. For example, they should develop assistive robots for medical rehabilitation, intelligent robots for education, picking and farming robots for agricultural production, etc. By expanding application scenarios, they can seize more market share and depend less on a single market.

Talent training and introduction:

EAI involves multiple disciplines, so OEMs and suppliers should recruit and train talents with interdisciplinary knowledge and skills. For example, Geely, BYD, Huawei and other companies established dedicated EAI research teams at the end of 2024, and they clearly require candidates to have knowledge in multiple fields such as machinery, automation, mechanics, computers, mathematics, electronic information and computing when recruiting talents. Companies can also improve the interdisciplinary capabilities of existing employees and build a professional talent team through internal training, cooperation with universities, etc.