Daily AI & Robotics Wrap: Humanoids Advance Amidst Expert Scrutiny and Security Concerns
The world of AI and robotics, particularly in the realm of humanoid development, continues its rapid evolution, marked this past day by significant technological releases, strategic industry collaborations, and crucial expert perspectives. While advancements promise more capable and integrated robots, critical voices and emerging security concerns highlight the complex path ahead for these human-like machines.
NVIDIA Unveils New AI Models and Simulation Tools to Accelerate Humanoid Robotics
NVIDIA has launched a suite of new robotics AI models and open-source simulation tools aimed at accelerating the research and development of humanoid robots. The announcement includes the opening of its Newton Physics Engine, Isaac GR00T foundation model, and Cosmos simulation libraries. These tools are designed to help developers create more adaptable and intelligent robot AI models.
The Newton Physics Engine, now available in Isaac Lab and managed by the Linux Foundation, is a GPU-accelerated, open-source platform co-developed with Google DeepMind and Disney Research. It is engineered to simulate complex scenarios crucial for humanoid robots, such as walking on varied terrain or handling delicate objects.
Alongside Newton, NVIDIA introduced Isaac GR00T N1.6, an open robot foundation model that incorporates Cosmos Reason. This vision-language-action model for physical AI helps robots interpret vague instructions and translate them into detailed action plans, incorporating common sense. Additionally, updated Cosmos world foundation models (WFMs) like Cosmos Predict 2.5 and Cosmos Transfer 2.5 aim to generate diverse datasets for large-scale AI training, enabling humanoids to perform more complex tasks with greater freedom. Robotics firms such as Franka Robotics, Techman Robot, LG Electronics, and Neura Robotics are already evaluating the GR00T models.
Expert Cautions Against Overstated Capabilities of Humanoid Robots in Homes
Despite recent impressive demonstrations, an expert from Northeastern University urges caution regarding the immediate prospects of humanoid robots in domestic settings. Ravinder Dahiya, a professor of electrical and computer engineering, commented on Google DeepMind’s recent showcase of Apptronik’s Apollo humanoid robot performing household tasks like folding laundry and sorting items using natural language commands.
DeepMind’s demonstration, powered by its Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 AI models, aimed to illustrate how large language models can enable physical robots to “perceive, plan, and think” for multi-step tasks. However, Dahiya emphasizes that while integrating vision technology with large language models is a significant step, it doesn’t equate to independent “thinking” by the robot. He notes that the robot’s capabilities are currently backed by extensive high-quality training data and structured algorithms.
Dahiya highlights that humanoid robots are still far from possessing human-level sensing or thinking capabilities, particularly in areas like tactile feedback, for which there is less training data compared to vision.
Pioneering Roboticist Rodney Brooks Dismisses Near-Term Human-Level Dexterity in Humanoids as ‘Pure Fantasy’
Renowned roboticist Rodney Brooks, co-founder of iRobot, has published a critical assessment of the current wave of humanoid robotics, stating that the idea of these robots achieving human-level dexterity in a meaningful timeframe is “pure fantasy thinking”. In his essay, Brooks argues that despite hundreds of millions, or even billions, in venture capital and tech company investments, today’s approaches will not deliver the required dexterity within decades.
Brooks frames dexterous manipulation as a core requirement for general-purpose humanoids to be economically viable. He points out that industrial grippers and suction tools remain the only robust options at scale, while human-like articulated hands have yet to demonstrate general dexterity or durability in real-world applications.
“In this post I explain why today’s humanoid robots will not learn how to be dexterous despite the hundreds of millions, or perhaps many billions of dollars, being donated by VCs (venture capitalists) and major tech companies to pay for their training.”
He suggests that much of the current optimism is based on unrealistic expectations about achieving parity with human hands in the near future.
Diligent Robotics Strengthens AI Advisory Board for Healthcare Humanoids
Diligent Robotics, a prominent company in humanoid robotics and embodied AI for healthcare, announced the expansion of its AI Advisory Board. The company has welcomed Siddhartha Srinivasa, Professor of Robotics at the University of Washington, and Zhaoyin Jia, a Distinguished Engineer specializing in robotic perception and autonomy.
The advisory board, initially launched in Fall 2024, aims to guide Diligent’s next generation of AI development and ensure its innovations are grounded in responsible and impactful practices. Srinivasa is recognized for his contributions to robotic manipulation and human-robot interaction, while Jia brings expertise in building large-scale AI systems for robots operating alongside people.
Andrea Thomaz, CEO and Co-Founder of Diligent Robotics, stated that the expanded board will combine the company’s success in real-world healthcare settings with the insights of world-class researchers to advance embodied AI and elevate standards for human-robot collaboration.
CJ Logistics Partners in State-Led Initiative to Develop Humanoid Robots for Logistics
South Korean logistics giant CJ Logistics has announced its participation in a state-led research initiative to develop humanoid robots equipped with dexterous robotic hands by 2028. This project, part of the Ministry of Trade, Industry and Energy’s “K-Humanoid Alliance,” aims to revolutionize logistics operations and secure a technological advantage in the sector.
The program, with a budget of 5.1 billion won ($3.7 million) including government funding, seeks to create robots capable of handling tasks requiring the precision of a human hand. Unlike traditional factory automation that excels at repetitive motions, logistics demands machines that can process a diverse range of goods varying in size, material, and shape.
AIDIN Robotics will lead the project, with CJ Logistics testing and validating the “AIDIN-Hand” humanoid robotic hand within its extensive logistics network. The collaboration also includes the Korea Electronics Technology Institute and Sungkyunkwan University. The initiative is expected to enable robots to perform more precise tasks, such as sorting and packaging, accelerating the commercialization of advanced humanoids in logistics.
Humanoid Robots Identified as Potential “Walking Trojan Horses” Due to Cybersecurity Flaws
A recent research paper published on arXiv has highlighted significant cybersecurity vulnerabilities in humanoid robots, warning that these advanced machines could become “walking Trojan horses” in the workplace. The study, conducted by Alias Robotics and independent security experts, focused on the Unitree G1 humanoid robot and uncovered alarming flaws.
The researchers found that even robots with what they considered “the most mature security architecture” could be compromised. Key vulnerabilities include a severe flaw in the robot’s Bluetooth Low Energy (BLE) provisioning protocol, allowing attackers within range to inject malicious commands and gain root access due to a hardcoded AES key shared across numerous devices.
Furthermore, the robot’s proprietary encryption for configuration files was found to have fundamental weaknesses, meaning compromising one device could compromise an entire fleet. Perhaps most concerning is that the Unitree G1 continuously streams vast amounts of data to servers in China without explicit notice or consent from operators, functioning as a persistent data collection device from startup.