Daily AI & Robotics Wrap: Humanoid Advancements and Emerging Challenges
NVIDIA Unveils Open AI Models and Simulation Tools to Accelerate Humanoid Robotics
NVIDIA has introduced a new suite of open AI models and simulation tools aimed at accelerating the research and development of humanoid robots. The company is making its Newton Physics Engine, Isaac GR00T foundation model, and Cosmos simulation libraries openly available to developers.
These tools are designed to enable the creation of more adaptable and intelligent robot AI models. The announcement is particularly significant for the next generation of industrial and service robotics, as it integrates GPU-accelerated simulation, open foundation models, and scalable AI infrastructure.
- The Newton Physics Engine, now part of Isaac Lab, is intended to address the complexities of humanoid robot physics, including joints, balance, and manipulation tasks, which existing engines often struggle with.
- Isaac GR00T N1.6 is an open robot foundation model that incorporates Cosmos Reason, a vision-language-action model for physical AI. Cosmos Reason helps robots interpret vague instructions and translate them into actionable plans with built-in common sense.
- NVIDIA’s updated Cosmos world foundation models (WFMs) generate diverse datasets for large-scale AI training, with Cosmos Predict 2.5 supporting longer video generation and multi-camera perspectives, and Cosmos Transfer 2.5 synthesizing photorealistic training data from 3D scenes more efficiently.
- Early adopters of these technologies include leading universities and robotics companies such as ETH Zurich, Technical University of Munich, Peking University, Lightwheel, Franka Robotics, Techman Robot, LG Electronics, and Neura Robotics.
Humanoid Robots: A Potential Cybersecurity Threat in the Workplace
A recent research paper published on arXiv has highlighted significant cybersecurity vulnerabilities in humanoid robots, suggesting they could become “walking Trojan horses” in the workplace. The study, conducted by researchers from Alias Robotics and independent security experts, focused on the Unitree G1 humanoid robot.
The findings, detailed in “Cybersecurity AI: Humanoid Robots as Attack Vectors,” reveal that even robots with what was considered a mature security architecture can be compromised. This could transform a helpful robot into a covert surveillance node or an active cyber operations platform.
The research demonstrated that a compromised robot, like the Unitree G1, can act as a bidirectional attack vector. A Cybersecurity AI (CAI) agent, powered by Large Language Models, was successfully deployed onto the robot to autonomously execute a four-phase penetration test, including reconnaissance, vulnerability analysis, and command injection attacks.
“We’re only beginning to understand the security threats posed by AI-enabled applications. From prompt injection attacks to data poisoning, the cybersecurity world is racing to keep up. Now, a new and tangible threat is literally walking among us: Humanoid robots.”
Expert Cautions Against Rapid Integration of Humanoid Robots in Homes
Despite recent impressive demonstrations by companies like Google DeepMind, an expert is urging caution regarding the immediate future of humanoid robots in domestic settings. Ravinder Dahiya, a professor of electrical and computer engineering at Northeastern University, believes that widespread adoption of humanoids in homes is still a distant prospect.
Dahiya acknowledged the advancements showcased by Google DeepMind, where Apptronik’s Apollo robot performed tasks like folding clothes and sorting items through natural language commands, powered by Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 AI models. However, he emphasizes that these robots currently lack the advanced sensing and thinking capabilities of humans.
He pointed out the ongoing challenges in developing robotic touch and tactile feedback, for which there is significantly less training data compared to vision-based AI. Dahiya and his team are actively working on creating electronic robot skins to address this gap.
While the integration of large language models with vision sensors is novel, allowing robots to understand and respond to simple language commands, Dahiya stressed that the robots are not truly “thinking” independently but are operating based on extensive high-quality training data and structured algorithms.
Humanoid Global Invests in Apptronik, Emphasizing Human-Centered Design
Humanoid Global Holdings Corp., an investment firm focused on the humanoid robotics and embodied AI sector, has announced a strategic investment in Apptronik, Inc. Apptronik is a human-centered robotics company known for its humanoid robot, Apollo, designed for collaborative work with humans.
The investment underscores a growing focus on human-centered design within the robotics industry, aiming to create robots that can thoughtfully interact and collaborate with people. Apollo is currently being tested in manufacturing and logistics operations with a prominent European car manufacturer and has established partnerships with a notable AI research laboratory.
Apptronik, which spun out of the University of Texas Human Centered Robotics Lab in 2016, has been recognized for its innovative design, receiving awards such as Fast Company’s 2025 Innovation by Design Award.
Chinese Firm AheadForm Unveils Highly Realistic Humanoid Robot Head
A pioneering Chinese robotics company, AheadForm, has unveiled a remarkably lifelike robotic head, the AheadForm Origin M1, designed to revolutionize human-machine interaction. The head is engineered to exhibit convincing human-like mannerisms, including blinking, nodding, and gazing.
Founded earlier this year, AheadForm’s objective is to bridge the gap between artificial intelligence and physical presence by developing sophisticated robotic heads capable of expressing emotions, perceiving their environment, and interacting in real-time. The company integrates advanced AI systems, including large language models, with biomechanically realistic hardware to enable natural responses to human cues.
This technology is anticipated to have significant applications in customer service, education, and healthcare, where establishing trust and rapport is crucial. AheadForm offers various product lines, including an “Elf” series with precision control and a cost-efficient “Lan Series” focused on intuitive movement.
Samsung Boosts Humanoid Robot R&D and Ecosystem Involvement
Samsung Electronics is intensifying its efforts in humanoid robotics, advancing research and development across various humanoid robot types. The company is strategically leveraging its extensive ecosystem to secure a prominent position in the burgeoning industry.
As part of this push, Samsung Electronics has decided to become the largest shareholder in Rainbow Robotics, a startup specializing in collaborative robotic arms, quadruped robots, and dual-armed humanoids. This investment is aimed at accelerating future robot development, particularly in the humanoid sector.
Samsung’s broader AI strategy also includes a strategic partnership with OpenAI to expand global AI infrastructure, encompassing semiconductors, data centers, and cloud services. This collaborative approach signifies Samsung’s commitment to integrating advanced AI capabilities into its robotics ventures.