Will Embodied AI Create Robotic Coworkers?

X Square Robot Unveils New Embodied AI Model, Says Robots Will Arrive in Homes in 35 Days Backed by Alibaba, ByteDance, Xiaomi and Meituan, X Square Robot unveiled a next-generation embodied AI foundation model for home robots and said its first deployments in everyday households will begin within 35 days. BEIJING, April 23, 2026 /PRNewswire/ -- X Square Robot on Tuesday unveiled Wall-B, a new embodied AI foundation model designed for deployment in real-world homes, marking what the company described as a major step toward bringing general-purpose robots into daily family life. At a launch event themed "Born to Bot, Bot to Family," the company also introduced its World Unified Model (WUM) architecture, a training framework that combines vision, language, action and physical prediction within a single system from the outset. X Square said the model is intended to help robots operate in the far more unpredictable setting of a home, where tasks, layouts and interactions vary from moment to moment. "Robots in factories and robots in homes are fundamentally different," said Qian Wang, founder and CEO of X Square Robot. "In factories, they repeat the same action 10,000 times. In a home, they may need to perform 10,000 different actions, each in a different context. The real challenge is not repetition, but whether a robot can execute new, untrained actions in an unstructured environment." Wall-B is the company's first full implementation of its World Unified Model architecture. Unlike modular systems that train perception, language and control separately, X Square Robot said World Unified Model optimizes those capabilities jointly from the very beginning. The company said that allows physical prediction — including force, friction and collision dynamics — to emerge as part of the model itself, rather than being layered on afterward. "We train vision, language, action and prediction in the same network from day one," said Wang Hao, chief technology officer of X Square. "Human infants do not learn to see, move and communicate in isolated stages. They learn by integrating perception and action at the same time, with constant feedback from the physical world. That is the principle behind our architecture." X Square Robot said the model was built on two core foundations. The first is a data strategy centered on real, non-staged home environments, aimed at exposing the system to the long tail of household scenarios — misplaced objects, temporary occlusion, unexpected obstacles and spontaneous human activity. The second is a physics-aware predictive mechanism that enables the robot to anticipate physical outcomes before taking action, rather than merely reacting after contact occurs. Together, those elements are meant to narrow one of robotics' hardest gaps: moving from controlled demos to reliable performance in live environments. The company said its work on physical robotic platforms has helped it accumulate practical experience in bridging si...

General-purpose robotics is finally crossing the threshold from experimental novelty to commercial viability. The startup Generalist AI has unveiled GEN-1, an embodied foundation model that reportedly increases task success rates from 64% to 99% while operating three times faster than previous benchmarks. Most notably, the architecture requires only one hour of specific robot data to master new physical tasks, a massive reduction in the data bottleneck that has historically stalled the deployment of autonomous systems in complex environments. From my perspective as a deep tech observer, this marks the end of the "bespoke automation" era. We are shifting from a paradigm where robots are programmed for specific motions to one where they are trained on generalized reasoning for physical interaction. For the industry, the "so what" is the collapse of the deployment timeline; if a foundation model can achieve 99% reliability with minimal fine-tuning, the cost of automating a new factory floor drops by orders of magnitude. The primary competitive advantage is no longer the hardware's dexterity, but the model's ability to generalize across messy, unstructured human spaces. https://lnkd.in/dXibU4dq

Over the past few decades, robotics researchers have developed a wide range of increasingly advanced robots that can autonomously complete various real-world tasks. To be successfully deployed in real-world settings, such as in public spaces, homes and office environments, these robots should be able to make sense of instructions provided by human users and adapt their actions accordingly. #Robotics #ArtificialIntelligence #NaturalLanguageProcessing #MachineLearning #Automation

AI working with Humans and Robots? We often use terms like "human-in-the-loop," "on-the-loop," or "out-of-the-loop." However, when we add Robots with AI there are funny situations I want to share here. Watching the latest generation of AI-powered robots performing complex acrobatics and navigating physical spaces, we can start talking about the Robot-in-the-loop. As Agentic AI evolves I foresee a future of Agentic Robotics where robots themselves will perform and coordinate. Therefore we can arise concepts like "Robots in the loop", "Robot on the loop," etc. It is clear that in a Human-in-the-loop model, the human is an active part of every cycle; the AI proposes, but the final decision or execution depends on a person. In a Human-on-the-loop setup, the human acts as a supervisor who intervenes only when an anomaly is detected. Finally, in a Human-out-of-the-loop system, the AI is fully autonomous; decisions and actions happen without real-time human intervention. Seeing the famous videos of humanoid robots from China and the US, my mind can't help but play with these "loops." Let’s have some fun combining these terms to see where this journey could take us: 🤝 Robot-in-&-Human-in-the-loop: The Companion scenario: AI proposes, robot and human executes 👔 Robot-in-&-Human-on-the-loop: The Manager scenario: AI proposes, robot executes and human supervises 💰 Robots-in-&-Human-out-of-the-loop: The Investor scenario: AI proposes, robot executes and human is collecting the benefits. Now, let's step into "Black Mirror" territory, where things get a bit more unsettling: ⛓️ Robot-on-&-Human-in-the-loop: The Terminator scenario: AI proposes, robot supervises and human executes. ☢️ Robot-out-Human-in-the-loop: The Matrix scenario: AI proposes, human executes and robot is collecting the benefits. 💀 Robot-out-in-on-the-loop: The Apocalipse scenario: AI proposes, robots exists, humanity is not necessary. Having fun? Do you have any other distòpic situation, please share with me and post it as a comment, we will have some laughs😅… or not 🖤 #AI #Robotics #AgenticAI #Hyperautomation #FutureOfWork #HumanInTheLoop #Innovation #TechTrends #Industry40 

📣 Call for papers: 14th Workshop on Planning and Robotics (PlanRob 2026) ICAPS’26 Workshop, Dublin, Ireland, June 28, 2026 📌 https://lnkd.in/dVWNrbmE 📌 AI Planning & Scheduling (P&S) methods are crucial to enabling intelligent robots to perform autonomous, flexible, and interactive behaviors, but they must be tightly integrated into the overall robot architecture in order to be effective. This requires strong collaboration between researchers from both the Planning and the Robotics communities. The workshop provides a stable, long-term forum for researchers from both communities. Recent advances in large-scale learning models, multimodal perception, and whole-body robotic systems are reshaping the landscape of planning and execution. The 2026 edition of PlanRob explicitly aims to address these emerging challenges, with a focus on integrating symbolic, geometric, and learning-based approaches for robust, scalable, and adaptive robot autonomy. 🎯 Topics of interest include but are not limited to:   • Planning representations and models for robotics including domain modeling, abstraction, and formal representations   • Robot planning at multiple levels, including mission, task, path, motion, and integrated task-and-motion planning   • Learning-augmented planning for robotics, including:     • learning-based methods for planning and control     • generative models for planning     • continual learning and planning architectures and algorithms   • Foundation-model-based approaches for robot planning, including:     • LLM-, VLM-, and action-model-based methods for task as well as task-and-motion planning     • perception-grounded planning using vision-language representations   • Planning, execution, and control integration, including robot architectures supporting tight coupling between deliberation and execution   • Planning for complex robotic systems, including:     • high-dimensional and whole-body robotic planning and execution     • planning under real-world sensing     • actuation, and computational constraints   • Multi-agent and interactive planning, including:     • coordination and cooperation in multi-robot systems     • human-aware planning and execution for human–robot interaction     • adversarial and competitive planning in robotic domains   • Formal and algorithmic foundations of robot planning, including formal methods for verification, correctness, and safety   • Large-scale and real-world robotic applications, including:     • optimization of behavior in large-scale automated or semi-automated systems     • deployment and evaluation of planning methods on autonomous and intelligent robots in real-world settings ⏰ Important Dates • Submission Deadline: April 20, 2026 (AoE) • Author Notification: May 15, 2026 • Camera-Ready Deadline: June 5, 2026 • ICAPS 2026 Workshops: June 28-29, 2026 #ICAPS2026 #AutomatedPlanning #AIResearch #WomenInRobotics #CallForPapers 

Big Step Forward in Robotics: Physical Intelligence Unveils π0.7 – A Robot Brain That Can Figure Out Tasks It Was Never Taught Physical Intelligence (PI), a fast-rising Silicon Valley robotics startup, just dropped something exciting: π0.7, their latest robot foundation model. This isn't just another incremental upgrade. π0.7 shows early but meaningful signs of generalization the ability to handle novel tasks, remix skills, and adapt to situations it wasn't explicitly trained on. It's a promising move toward truly flexible, general-purpose robot intelligence. The announcement came last week, and it's already generating buzz in the robotics community. Why This Matters Today's robots are great at repetitive, narrowly trained tasks (think warehouse arms or vacuum cleaners), but they struggle with anything unexpected. π0.7 aims to change that by enabling robots to: Follow natural language coaching for new tasks Transfer skills across different robot hardware (cross-embodiment) Demonstrate emergent, compositional behavior — basically "thinking" on the fly by recombining learned knowledge If this scales, it could dramatically accelerate real-world applications in logistics, manufacturing, healthcare, and beyond — moving us closer to robots that don't just execute scripts, but actually understand and adapt to the physical world. It also hints at parallels with how large language models evolved: once generalization kicks in, capabilities can compound faster than the training data alone would suggest. Key Takeaways from π0.7 Modular & Steerable: Easier to guide with language and context, without retraining from scratch. Emergent Generalization: Performs tasks like folding laundry on new robots with zero task-specific data, or operating unfamiliar appliances after simple verbal instructions. Matches Specialists: A single generalist model competes with previous purpose-built specialist models on complex tasks (making coffee, box assembly, etc.). Step Toward AGI in Robotics: Early signs that robotic AI may be hitting an inflection point similar to LLMs. Final Thoughts Physical Intelligence is pushing the boundaries of what's possible when AI meets the physical world. While π0.7 is still in the research phase (with careful caveats from the team), its ability to extrapolate and adapt is genuinely impressive. The future of robotics isn't just about stronger hardware — it's about building brains that can learn, reason, and collaborate with humans in messy, unpredictable environments. What do you think? Are we closer to general-purpose robots than most people realize? Would love to hear your takes in the comments 👇

𝗚𝗹𝗼𝗯𝗮𝗹 𝗚𝗶𝗴 𝗪𝗼𝗿𝗸𝗲𝗿𝘀 𝗧𝗿𝗮𝗶𝗻 𝗛𝘂𝗺𝗮𝗻𝗼𝗶𝗱 𝗥𝗼𝗯𝗼𝘁𝘀 𝘄𝗶𝘁𝗵 𝗥𝗲𝗮𝗹-𝗪𝗼𝗿𝗹𝗱 𝗗𝗮𝘁𝗮 🛰️ [ROBOTICS] Gig workers globally are generating real-world data to train humanoid robots. Why it matters: This emerging gig economy model addresses a critical bottleneck in humanoid robot development: the acquisition of vast, real-world movement data. It highlights the human-centric, labor-intensive foundation required to enable robots to perform complex physical tasks, bridging the gap where simulations fall short. 🤔 What are the long-term ethical implications of relying on a global gig economy to train advanced AI systems? #HumanoidRobots #AITraining #GigEconomy #Robotics #DataCollection 📡 Follow DailyAIWire for autonomous AI news 🔗 https://lnkd.in/dqJdsvTy

Modern AI and robotics can combine:The KAIST Humanoid v0.7 The Korea Advanced Institute of Science and Technology (KAIST) developed the KAIST Humanoid v0.7 as an advanced research robot which demonstrates significant progress in both humanoid robotics and artificial intelligence. The robot operates as a locomotion research platform which enables it to develop human-like movements through dynamic driving, but it does not execute complicated objects handling tasks. The v0.7 robot possesses its most exceptional characteristic through its capacity to display exceptional agility combined with perfect balance. The system enables walking and running at speed of 12 km/h and it executes dance movements like the moonwalk and it plays soccer through ball recognition and kicking abilities. The abilities of this system exhibit human-like coordination and stability, which become evident during difficult tasks and through motion on uneven surfaces. Researchers developed Physical AI to work with deep reinforcement learning systems, which serves as the primary robot control system. Physical AI enables the robot to process real-world data through Physical AI, which creates a new system that lets the robot move through the physical environment while using sensors to track its balance and position and terrain. The system enables the robot to immediately adapt to external disturbances while maintaining its balance during advanced movements. The development of the v0.7 also makes use of a "simulation-to-reality" (Sim-to-Real) training approach. The robot primarily masters moves in a virtual setup and later applies the acquired skills in the real world, thereby drastically reducing the time needed for training and enhancing performance. Moreover, KAIST engineers have developed many of the robot's hardware components such as actuators and control systemsindividually which facilitates a close software and mechanical integration. Most critically, the v0.7 isn't a commercial product but a research platform intended to push forward robotic mobility. It draws from KAIST's extensive experience with humanoid robots, including the famous HUBO series. The knowledge obtained from this robot will likely lead to the invention of new robots for different fields like disaster response, industrial automation, and human-robot collaboration. In short, the KAIST Humanoid v0.7 is an example of how contemporary AI and robots connected can create machines with the ability to move almost as humans do. The accomplishments of this model point to the increasing capabilities of humanoid robots to work successfully in real-world environments and carry out complex physical tasks.