Humanoid Robots Speak at Web Summit as Agentic Robotics Reshapes Industry

The robotics industry is experiencing an unprecedented moment in November 2026. For the first time in history, humanoid robots have been registered as official speakers at Web Summit in Lisbon, one of the world’s largest technology conferences. This milestone represents more than a publicity stunt — it signals a fundamental shift in how robots are integrating into human social and professional spaces.

Humanoid Robots Take the Stage at Web Summit

At this year’s Web Summit, four Chinese-made humanoid robots were officially registered as speakers for the event. Among them, Agibot’s representative William Shi delivered a presentation that unexpectedly turned competitive, criticizing rival robotics companies in what can only be described as the first instance of robot-on-robot trash talk at a major tech conference.

Shi directed pointed criticism at Tesla, Figure AI, and Unitree. He claimed Tesla relies on carefully produced videos rather than public live demonstrations. He noted that Figure AI robots never dance. And he observed that Unitree robots only dance and never actually work. In contrast, Shi positioned Agibot’s robots as versatile machines that both dance and perform real factory work.

While the competitive jabs drew attention, the deeper significance lies in the fact that humanoid robots have reached a level of capability and reliability where they can participate meaningfully in human professional gatherings. This is not a remote-controlled prop on stage — these are autonomous machines capable of interaction, response, and presentation.

The Competitive Landscape Intensifies

The Web Summit moment highlights an increasingly crowded and competitive humanoid robotics market. Companies from the United States, China, and Europe are racing to develop robots that can operate in factory environments, warehouses, and eventually homes.

• Tesla continues developing its Optimus platform, though critics note the company’s reliance on demonstration videos over live public showings
• Figure AI focuses on practical task execution but faces legal scrutiny after a lawsuit alleged its robots could cause serious injury
• Unitree has gained viral attention for its robots’ agility and entertainment capabilities, including dancing demonstrations
• Agibot positions itself as the versatile competitor, combining practical industrial deployment with public demonstration capabilities
• Boston Dynamics recently launched Stretch Gen 2, a next-generation autonomous mobile robot designed for trailer unloading in logistics operations

The competitive tension reflects a market that has moved beyond laboratory demonstrations into real commercial deployment. Humanoid robot shipments surged 300 percent in 2026, and companies are now racing to prove their robots can handle genuine industrial workloads, not just curated demonstrations.

Agentic Robotics: A Paradigm Shift in Robot Intelligence

While the Web Summit showcases the social arrival of humanoid robots, a parallel revolution is transforming how robots learn and improve. Researchers at UC Berkeley, in collaboration with NVIDIA, have unveiled what they call agentic robotics — an approach that uses AI agents to generate, test, and refine robot-control programs autonomously.

Ken Goldberg, a UC Berkeley roboticist recently inducted into the National Academy of Engineering, described the approach as a fundamental shift. Traditional robotics required either extensive manual programming or enormous datasets for imitation learning. Agentic robotics offers a third path: AI agents that write robot code, test it in simulation, identify failures, and iteratively improve the program without human intervention.

The results have been striking. Goldberg’s team gave OpenAI’s Astra model access to a simulation interface and a short video of a robot wiping ink off a metal bar. In less than an hour, the AI created a physics simulation that accurately reproduced the video, including realistic deformations of the sponge. The AI then generated a robot-control program, tested it in simulation, and when researchers ran the resulting code on a real robot, it worked.

When the robot initially failed — pushing over the metal bar because it did not correctly infer the bar’s mass — the AI used video of the failed attempt to improve its simulation and revise the control program. This self-improvement loop is what Goldberg described as giving him goosebumps, comparing the moment to the public release of the World Wide Web in 1993 and the debut of ChatGPT in 2022.

Real-World Impact Already Underway

The agentic robotics approach is not confined to academic laboratories. Ambi Robotics, a Berkeley-based company specializing in logistics and warehouse fulfillment, has already applied the methods to its production fleet. By converting production data into robot behavior updates, the company increased sorting capacity by over 15,000 packages annually per robot. The improvement is now deployed across 30 percent of Ambi’s U.S. fleet.

This deployment demonstrates a critical advantage of the agentic approach: robots that can self-improve based on their own operational data. Rather than requiring engineers to manually retune behavior for every new scenario, the AI agents analyze performance data, identify inefficiencies, and generate improved control programs that human engineers can review and approve.

The Data Challenge Remains

Despite the progress, significant challenges persist. Goldberg previously warned of a 100,000-year data gap — the notion that while AI language models were trained on vast amounts of internet text and images, there is no comparable data source for training robots on physical world interactions. Every door handle, every object weight, every surface texture represents a unique interaction that robots must learn to navigate.

Agentic robotics offers a way around this data bottleneck by leveraging simulation environments and AI reasoning capabilities. Rather than requiring millions of physical demonstrations, AI agents can generate and test thousands of variations in simulation, learning from both successes and failures at a pace impossible with physical robots alone.

The approach also addresses another persistent challenge: the cost of data collection. The current competitive race in humanoid robotics is less about the physical robot body and more about the cost and efficiency of acquiring the training data that makes robots useful. Companies that can generate, test, and refine robot behaviors quickly and inexpensively will hold a decisive advantage.

Industrial Applications Lead the Way

Goldberg cautioned against expecting humanoid robots to fold laundry at home in the immediate future. The near-term impact of agentic robotics will be felt in industrial settings where tasks are repetitive, measurable, and operate at scale. Warehouse sorting, manufacturing assembly, and logistics operations are the primary beneficiaries.

Boston Dynamics’ Stretch Gen 2 exemplifies this industrial focus. Developed over two and a half years, the robot is specifically designed for autonomous trailer unloading — a physically demanding task that has been difficult to automate. The second generation focuses on reliability and consistent performance, addressing the real-world deployment challenges that first-generation robots often face.

Other industrial developments include Foundry Robotics raising $100 million in Series A funding to build autonomous factories for defense and industrial hardware assembly. Walmart expanded its agreement with Symbotic to deploy automation across 25 U.S. regional distribution centers. These investments signal that the market for industrial robotics is maturing rapidly.

Geopolitical Dimensions of the Robotics Race

The robotics competition carries significant geopolitical implications. Chinese regulators recently blocked SoftBank’s proposed $5.4 billion acquisition of ABB Robotics, reflecting growing tensions around who controls critical robotics and automation technology. The United States has also moved to restrict certain Chinese humanoid robots, citing security concerns.

Meanwhile, Chinese manufacturers are achieving production milestones at remarkable speed. In Guian New District, a facility produced the first 100 humanoid robot chassis in just 35 days, reaching mass production capability. China’s humanoid robot ecosystem spans from component manufacturing to finished platforms, creating an integrated supply chain that Western competitors are still building.

The Web Summit appearance of four Chinese humanoid robots as official speakers underscores China’s growing confidence and visibility in the global robotics arena. The competitive dynamic between American, Chinese, and European robotics companies will shape the industry’s trajectory for years to come.

Looking Forward

The convergence of agentic robotics, mass production capabilities, and competitive market dynamics points to a transformative period ahead. Robots that can write and improve their own code, backed by companies investing billions in development and deployment, are moving from novelty to necessity in industrial operations.

The open-source release of Berkeley and NVIDIA’s agentic robotics code, along with an industry-standard benchmark for evaluating robot performance, will accelerate adoption across the field. As more researchers and companies experiment with the approach, the pace of improvement will likely accelerate beyond what any single company could achieve alone.

For now, the headlines may focus on robot trash talk at tech conferences. But the real story is the emergence of robots that learn, adapt, and improve themselves — a development that may ultimately prove as significant as the internet and AI language models that preceded it.

---

Edited by Palawan @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous