A video of two humanoid robots trading punches and kicks in a Bay Area storefront has gone viral across social media platforms, billed by organizers as the first public fight between a Unitree robot and an EngineAI model. The footage, shared on X by virtual-reality innovator Cix Liv, shows two bipedal robots in a makeshift ring with a human referee, while a small crowd cheers as the machines punch, dodge, and circle each other.

The caption described it as “the first fight ever between an Engine and a Unitree robot at our new store space in SF”. The video accumulated thousands of views and likes across platforms, with viewer reaction ranging from amazement to practical frustration – one commenter noting they would prefer robots that could load a dishwasher over ones that can fight.

The Two Platforms

Unitree’s G1 humanoid, priced at approximately $16,000, has become one of the more commercially accessible full-size humanoid platforms available. The company drew global attention after its robot won the world’s first robot combat competition last year, and the G1 is now being deployed across contexts ranging from Haneda Airport’s baggage handling trial to Seoul’s Jogyesa Temple ordination ceremony. Unitree also showed its H2 humanoid at CES 2026.

EngineAI’s T800, which debuted at CES 2026, stands 1.73 meters tall and weighs 75 kilograms. The company has positioned it around stability and durability, with a starting price of $25,000 and first shipments scheduled for mid-2026.

What the Fight Actually Shows

Several commenters noted that the robots in the video appear to be remotely controlled rather than fighting autonomously – a distinction that matters for assessing what the footage demonstrates about AI capability versus hardware performance. Remote-controlled robot combat has existed for decades in formats like BattleBots. What is new here is the use of full-size bipedal humanoid platforms in an unstructured public environment, with human-scale proportions and movement.

The viral reach of the clip reflects both genuine public fascination with humanoid robot capability and the role of social media in amplifying robot demonstrations that would previously have been confined to trade show floors and research facilities. Whether robot combat develops into a structured commercial format – as Unitree’s earlier competition win suggests it might – or remains a demonstration category will depend on whether organizers can build repeatable event infrastructure around it.

For both companies, the attention generated by a single informal storefront fight represents free market validation that the hardware is visually compelling enough to capture mass audiences without a formal launch event.

Researchers at the Italian Institute of Technology in Genoa have developed a soft robotic arm that moves like an elephant’s trunk, capable of elongating, compressing, bending, pinching, scooping, and reaching – all using a single continuous structure with no hard joints or separate gripper. The prototype emerges from PROBOSCIS, a five-year EU-funded research initiative bringing together biologists, engineers, and materials scientists to decode the mechanics of elephant trunk movement and translate them into robotic hardware.

The project was initiated by Lucia Beccai, a soft robotics researcher at IIT, who identified the elephant trunk as the closest natural analog to what robotics has been missing: a single structure that can handle a grape and lift a heavy load with the same physical system.

Why the Trunk Is Difficult to Replicate

The elephant trunk is a muscular hydrostat – the same structural category as an octopus tentacle or a human tongue. It contains more than 100,000 individual muscles and no skeleton, allowing it to extend, contract, bend, and twist in any direction simultaneously. Critically, there is no distinction between arm and gripper: the entire trunk is one continuous structure capable of whole-body manipulation.

To study trunk mechanics, Professor Michel Milinkovitch at the University of Geneva led a team using motion-capture techniques borrowed from film production – placing reflective marker spots on elephant trunks and recording movement with high-speed cameras. The analysis revealed that elephants combine a small set of fundamental behaviors – shortening, elongating, and bending different sections – to achieve complex manipulation tasks. One particularly unexpected finding was that when reaching behind their heads, elephants create temporary pseudo-joints by stiffening sections of the trunk, effectively generating a shoulder and elbow structure on demand.

The Engineering Solution

Beccai’s team translated these biological findings into a 3D-printed soft robotic system. The prototype uses pneumatic actuators – balloon-like structures that extend and contract as they are inflated and deflated with air – combined with a mesh-like lattice structure that can deform in multiple directions. The entire device, including optical sensors that provide real-time feedback on touch and bending, is printed in a single continuous process from the same soft resin.

The single-material approach is central to the design’s performance. By eliminating the material and mechanical interfaces between components, the system achieves continuity of motion combined with integrated sensory feedback – a combination that conventional rigid-arm robotics with separate sensing systems cannot replicate.

The control architecture draws on Milinkovitch’s biological finding about muscle synergies. Rather than controlling each actuator individually, the system manages a small number of coordinated muscle group patterns, with the physical structure handling the mechanical complexity. This approach reduces computational demand and energy consumption, making battery-powered deployment outside laboratory conditions more viable.

Applications and Limitations

The research project wrapped up in April 2025, and the prototype remains a laboratory demonstrator. The team says it already addresses most of the design constraints limiting today’s robotic arms in unstructured environments. Potential applications identified by the researchers include soft fruit harvesting, domestic tasks such as sorting laundry and handling fragile dishes, environmental operations in fragile ecosystems, and search and rescue in rubble environments where a soft arm can navigate gaps and use touch sensing to locate people.

Beccai’s primary target is assistive robotics. “My dream is to build a system in healthcare that can help a disabled or elderly person by lifting them, but at the same time hand over a fork or a fresh piece of fruit,” she said. A robot strong enough for patient transfers yet gentle enough for daily object handling could meaningfully extend independent living for people with mobility limitations – and its softness, she argues, means it need not feel intimidating to be around.

UBTECH Robotics and Hitachi China have signed a strategic cooperation agreement to jointly develop and deploy humanoid robot applications across Hitachi’s manufacturing operations in China. The partnership focuses on sectors where Hitachi holds established market positions – elevators, building systems, healthcare, industrial equipment, and semiconductor manufacturing equipment – and is already in active deployment testing with UBTECH’s Walker S2 industrial humanoid.

The agreement moves beyond a memorandum of understanding stage. Both companies confirmed that cooperation has entered what they describe as the substantive verification stage, with Walker S2 units introduced into Hitachi manufacturing scenarios and on-site operation testing underway.

Elevator Manufacturing as the Initial Focus

The most developed application is at Hitachi Elevator’s manufacturing site, where Hitachi China Research Institute, UBTECH Robotics, and Hitachi Elevator are conducting secondary development of humanoid robot applications. The work targets full-process flexible intelligent flow and data integration across key elevator production processes – tasks that require the high precision and adaptability that fixed automation has historically struggled to deliver in low-volume, high-mix manufacturing environments.

Elevator manufacturing presents a relevant test case for humanoid deployment. The components are large, varied, and assembled in configurations that change between product types, creating the kind of flexible production demand that humanoid robots are specifically designed to address. Successfully automating key processes here would provide a replicable model for Hitachi’s other manufacturing segments.

The Broader Scope

The agreement covers the full range of Hitachi Group’s competitive markets in China. Healthcare and semiconductor manufacturing equipment represent the other near-term targets alongside elevators – both sectors where precision requirements are high, labor costs are rising, and production flexibility is commercially valuable.

UBTECH brings Walker S2 hardware and its software stack for industrial humanoid deployment. Hitachi brings system integration experience, existing manufacturing infrastructure, and a customer base across the sectors the partnership is targeting. The combination follows a pattern emerging across the humanoid industry: established industrial operators providing the deployment context and domain knowledge, while robotics companies provide the hardware and AI capability.

The partnership also extends UBTECH’s international footprint beyond its existing European presence through Terra Robotics in the DACH region, adding a major Japanese industrial partner with operations across Asia to its commercial network.

Unitree Robotics has officially launched UniStore, a task and motion application store for its robot lineup, enabling users to download and install new robot capabilities directly from a smartphone app. The platform covers Unitree’s G1 and H1 humanoid robots, the B2 quadruped, and the Go2 robot dog. The company describes it as the world’s first humanoid robot application store.

The launch introduces a consumer-facing software distribution model to humanoid robotics – one that mirrors the UX structure of smartphone app stores but applies it to physical robot behavior.

What UniStore Offers at Launch

UniStore opens with four core modules: User Square, a community and discovery layer; Motion Library, a catalog of downloadable execution actions; Dataset, a repository of robot interaction data; and Developer Center, a third-party publishing environment for developers building new motion and task packages.

The Motion Library launches with 24 available actions, currently offered on a limited-time free trial. These include Jackson-style dance moves, Charleston, jump-dancing, cheering sequences, Mantis Boxing, and a category the company describes as meme or “整活” actions – playful, culturally referential motion packages designed for entertainment and social sharing. Users install packages directly from the phone app with a single tap.

The Platform and Ecosystem Strategy

The Developer Center is the structurally significant component. By opening a publishing pathway for third-party developers, Unitree is building the infrastructure for an external ecosystem to grow around its hardware – a model that has proven durable in smartphones, gaming consoles, and enterprise software, but has not previously been applied to consumer and prosumer robotics at this scale.

The implication is a platform dynamic: as Unitree’s installed base of G1, H1, B2, and Go2 units grows, the addressable audience for third-party developers increases, which attracts more developers, which increases the platform’s value to hardware buyers. The Motion Library at launch is primarily entertainment-oriented, but the Developer Center is designed to support the full range of task and motion packages a developer might build – including functional applications for logistics, inspection, and service environments as the hardware deployment base matures.

The launch positions Unitree in contrast to proprietary robotics platforms where capabilities are controlled by the manufacturer. It also arrives as Unitree is preparing for a Shanghai IPO targeting a valuation of up to $7 billion, filed in March, making the ecosystem announcement a relevant signal to potential investors about the company’s platform ambitions beyond hardware sales.