Physical AI News (June 9, 2026 Issue)
Update Date: 2026/6/9
Executive Summary
On June 8, 2026, a notable trend emerged where Physical AI, centered around NVIDIA, is simultaneously expanding into AI factories, robotics, mobility, semiconductors, and power infrastructure. LG, Hyundai, SK hynix, and Doosan have all presented visions that integrate not only computing foundations but also manufacturing sites, logistics, construction machinery, and data centers, as well as power supplies. Meanwhile, cases from TARS and GS E&C demonstrate the concretization of field implementation, such as tactile hands and construction robots. In terms of research, beyond the universal theory of VLA and world models, the reorganization of data, embodiment, physical reasoning, and reward design is progressing.
*The created article content was input into Gamma to automatically generate slides. If you find slides easier to view, please take a look here.
1️⃣ LG × NVIDIA "M.A.P.": Integrating Physical AI, AI Infra, and Mobility
📎 Source: PRNewswire / NVIDIA Blog
Under the "M.A.P." initiative, which bundles Physical AI, AI infrastructure, and mobility, LG and NVIDIA have expanded their cross-group partnership with AI factories at the core. LG CNS is integrating Isaac, Cosmos, and Isaac GR00T into PhysicalWorks to accelerate the introduction of AI robots for manufacturing and logistics sites. LG Electronics is responsible for cooling and modular AI infrastructure design aligned with DSX, LG Energy Solution for 800V DC power solutions, and LG Uplus for large-scale AI data centers supporting Rubin GPUs, with explicit mention of collaboration with the DRIVE Hyperion/DRIVE platform.
2️⃣ Hyundai Motor Group × NVIDIA: Expanding from mobility to robotics / AI factories
📎 Source: Yonhap News Agency / Hyundai Motor Group
At a meeting at their Seoul headquarters, Hyundai Motor Group and NVIDIA discussed expanding collaboration in AI and future mobility, with CEO Jensen Huang expressing an intention to comprehensively cooperate from mobility to robotics and AI factories. Meanwhile, Hyundai has separately announced a plan to invest approximately 9 trillion won with the South Korean government and North Jeolla Province to develop an innovation hub in Saemangeum consisting of an AI data center, robot manufacturing cluster, PEM water electrolysis plant, solar infrastructure, and AI hydrogen smart city. While the possibility of future NVIDIA participation is mentioned, a concrete scheme to integrate factories, robots, and data centers into the same Physical AI foundation has not been shown.
3️⃣ NVIDIA × SK hynix: Multi-year partnership supporting computing foundations for Physical AI and autonomous fabs
📎 Source: NVIDIA Newsroom / Reuters
SK hynix and NVIDIA announced a multi-year technical partnership for next-generation memory in line with NVIDIA's AI infrastructure roadmap, agreeing to strengthen the supply of advanced memory for AI factories. SK hynix is diversifying into new markets for AI infrastructure, personal AI, and physical AI, and will co-develop memory for the Vera Rubin AI supercomputer, Vera CPU, RTX Spark-equipped PCs, and Jetson Thor robotics. Additionally, both companies will accelerate semiconductor simulation and TCAD using CUDA-X and PhysicsNeMo, and advance the construction of digital twins and autonomous operations of fabs utilizing Omniverse, OpenUSD, and cuOpt.
4️⃣ Doosan Group × NVIDIA: Agentic Robot OS and Physical AI expansion to construction and logistics equipment
📎 Source: NVIDIA Blog / Reuters
Doosan Group and NVIDIA have agreed to expand collaboration in physical AI and AI factory infrastructure across Doosan Robotics, Doosan Bobcat, Doosan Enerbility, and Doosan Electro-Materials BG. Doosan Robotics is integrating Isaac Sim/Isaac Lab, Cosmos, Newton, and Jetson Thor to promote an Agentic Robot OS that connects perception, reasoning, simulation, learning, and on-device inference, and is also developing high-value-added tasks such as depalletizing and polishing, as well as new form factors such as dual-arm and humanoid robots. Bobcat is considering the application of NVIDIA physical AI technology to construction, landscaping, agriculture, and material handling equipment, while Enerbility is exploring power solutions for AI factories using gas turbines, SMRs, and hydrogen fuel cells. Electro-Materials will support next-generation data centers with high-performance CCL for the MGX ecosystem.
5️⃣ TARS DexHand / AWE 3.0: Strengthening "physical understanding of fingertips" with a 21-degree-of-freedom tactile hand
📎 Source: EQS News / PRNewswire UK
At ICRA 2026, TARS showcased real-world embodied AI with DexHand and AWE 3.0. DexHand features a 21-degree-of-freedom configuration that reproduces human metacarpal and phalangeal topology 1:1, along with joints equipped with high-precision reducers, demonstrating high tracking and operability through 26 English alphabet sign language gestures and real-time mirror control demos. The fingertips contain ultra-compact cameras that capture 0.05mm-class textures at over 240Hz, and AWE 3.0 and SenseHub understand and predict physical properties such as hardness, roughness, and slip risk. With a highly rigid quasi-direct drive design limited to three types of motors and reducers, mass production capability with an eye toward automated assembly lines is also strongly emphasized.
6️⃣ GS E&C × Daedong Robotics: Joint R&D for AI autonomous robots for construction sites
📎 Source: The Asia Business Daily / The Korea Herald
GS Engineering & Construction and Daedong Robotics have signed a partnership for the joint development of AI autonomous robots for construction site automation and smart construction technology. Both companies will conduct field trials of existing mass-produced AI field robots at construction sites, proceeding with applications from areas that are easy to automate, such as material transport and repetitive tasks. Daedong is responsible for design, R&D, and PoC utilizing its autonomous robot platform, while GS E&C is responsible for demonstration fields, site requirement definition, and verification and commercialization support, aiming to develop dedicated robots optimized for highly changeable construction sites.
7️⃣ arXiv cs.RO New Arrivals: Research trends questioning VLA and world model dependence
📎 Source: arXiv cs.RO new listings
In the June 8, 2026, cs.RO new arrivals, a position paper was highlighted that criticizes the stance of reducing general-purpose robot intelligence to VLA and world model scaling problems, and presents four research tasks: "data interfaces" that extract useful supervisory signals for robots from human motion, internet videos, and simulations; "embodiment interfaces" that map human motion to robot behavior; "world model interfaces" that perform physics-based 3D reasoning; and "reward interfaces" that estimate task progress and success from video and language. Additionally, PhyRoGen is a framework that automatically generates combination puzzles of interdependent objects through procedural content generation, showing that it can generate 24 physical puzzles from 6 types of generators, solve them in 1 to 300 seconds with sampling-based planning, and that all puzzles are manipulatable in a KUKA LBR iiwa simulation, demonstrating its utility as a benchmark for evaluating manipulation foundation models and synthetic data augmentation.
Comprehensive Discussion
The key takeaway from the topics on June 8, 2026, is that Physical AI is shifting from a performance competition of individual robots to an industrial infrastructure competition that integrates AI infrastructure, semiconductor memory, power, simulation, and field operations. In particular, NVIDIA is solidifying its position as a core platform that supports the development, training, inference, and operation of robots through its Isaac, Cosmos, Omniverse, Jetson, and GPU roadmaps. On the other hand, as shown by the cases of Hyundai and construction robots, challenges such as field requirements, ROI, safety, and commercialization schemes remain between conception and implementation. Research-based problem statements indicate that for intelligence to truly generalize in the physical world, embodiment and environmental understanding are essential, not just model scale.
Points to watch in the future
The focus of NVIDIA partnerships is shifting from GPU supply to becoming an industrial OS that bundles robots, factories, data centers, and power sources, making the struggle for leadership among companies critical.
Companies like LG and Doosan that discuss AI factories and robot implementation simultaneously are more likely to move beyond PoC and closer to cross-divisional mass production deployment.
The partnership with SK hynix shows that the competitiveness of Physical AI depends not only on the robot itself but also on HBM, next-generation memory, and fab autonomy.
The tactile technology of TARS DexHand is attracting attention as a turning point from vision-centric robot perception to contact intelligence that reads hardness, roughness, and slippage.
As the arXiv discussions suggest, scaling VLAs and world models alone is insufficient; research that connects human motion, physical differences, physical reasoning, and reward design will be the key.
