A live warehouse trial involving SAP, robotics firm Humanoid, and automotive supplier Martur Fompak is offering a clearer view of how enterprise software systems may begin to directly control physical labor. The project connects SAP’s AI agent framework with a mobile humanoid robot, enabling task execution that originates from business logic rather than pre-programmed robotic routines.

The result is less about a single robot completing a task and more about a shift in system architecture. For decades, enterprise software has optimized operations in the digital layer, while robots have executed fixed workflows on the factory floor. This trial suggests those layers are beginning to merge.

From Enterprise Decisions to Physical Actions

At the center of the integration is a connection between SAP’s Joule AI agent and Humanoid’s KinetIQ robotics stack. In the proof of concept, a cloud-based software agent generated instructions based on operational needs and dispatched them directly to a physical robot.

The HMND 01 Alpha Wheeled robot then executed a logistics workflow inside a warehouse environment. It autonomously navigated to a designated pallet, retrieved a standardized container, and delivered it to a trolley as part of an order fulfillment cycle. The robot repeated the process across multiple iterations, handling different container types and operating within existing facility constraints.

This model differs from traditional automation, where tasks are predefined and robots follow rigid sequences. Instead, the system treats the robot as an endpoint of enterprise decision-making, allowing workflows to adapt dynamically based on business inputs.

SAP described the approach as embedding business context into robotic behavior, enabling machines to respond to operational priorities in real time rather than executing isolated tasks.

Embodied AI Moves into Enterprise Infrastructure

The trial reflects a broader convergence between two previously separate domains: enterprise AI and embodied AI. Software agents have long been capable of making decisions across functions such as procurement, scheduling, and inventory management. What has been missing is a direct link between those decisions and physical execution.

By connecting an enterprise agent to a humanoid system, the project demonstrates how digital instructions can translate into real-world actions without human mediation. In this case, warehouse picking becomes an extension of enterprise planning systems rather than a standalone robotic function.

The robot operated with a dual-arm payload capacity of 8 kilograms and was tested under real operational conditions over a multi-week deployment. The structured rollout included simulation and physical twin development, followed by on-site setup, training, and optimization, reflecting a methodology closer to enterprise software deployment than traditional robotics testing.

Executives involved in the project framed the outcome as a transition point. The ability to integrate robots into enterprise systems, measure them against operational metrics, and deploy them within existing workflows suggests that humanoid platforms are moving beyond experimental pilots.

A Bridge Between Pilots and Scaled Deployment

Despite the progress, the trial remains a proof of concept. The next phase will focus on extended validation in live production environments, where reliability, uptime, and integration complexity become more visible.

Still, the implications are notable. If enterprise systems can reliably orchestrate fleets of robots as extensions of digital processes, automation could shift from task-specific deployments to system-wide coordination. In such a model, robots are not programmed individually but managed as part of a broader computational infrastructure.

For Humanoid, the project provides evidence that its platform can operate within industrial constraints. For SAP, it demonstrates a pathway to extend enterprise AI beyond software into physical operations.

More broadly, the trial highlights an emerging pattern in robotics: the value of a robot is increasingly tied not just to its hardware capabilities, but to how seamlessly it integrates into existing digital systems. As that integration improves, the boundary between decision-making and execution may continue to narrow.

Xiaomi has introduced a redesigned bionic hand for its CyberOne humanoid robot, marking a step forward in one of the most technically challenging areas of robotics: human-level manipulation. The upgrade combines mechanical redesign, tactile sensing, and thermal management, reflecting how leading robotics programs are shifting from demonstration toward operational capability.

While much of the humanoid robotics conversation has focused on locomotion and general intelligence, manipulation remains a bottleneck for real-world deployment. Xiaomi’s latest iteration suggests that progress is increasingly coming from integrated system improvements rather than single breakthroughs.

Human Scale Design Meets Industrial Precision

A central change is the reduction of the hand’s size by roughly 60%, bringing it closer to the proportions of a human hand. This shift is not cosmetic. Matching human-scale geometry allows robots to interact more naturally with existing tools, components, and environments that were not designed for machines.

At the same time, Xiaomi has expanded the hand’s degrees of freedom, enabling more precise articulation and grip control. The result is a system better suited for tasks that require fine motor skills, such as component handling or in-hand manipulation.

This aligns with a broader industry direction. Rather than redesigning factories around robots, companies are increasingly trying to build robots that can operate within human-centered infrastructure. Achieving this requires not only mobility, but also dexterity that approaches human capability.

Tactile Sensing and Data as a Core Layer

Beyond mechanical improvements, Xiaomi has significantly expanded tactile sensing across the hand. The system now covers approximately 8,200 mm² across fingertips, finger pads, and the palm, allowing the robot to detect pressure and contact in a more distributed and nuanced way.

This is particularly relevant in scenarios where vision alone is insufficient. Occlusions, variable lighting, and complex object interactions often limit camera-based perception. Tactile feedback provides an additional channel for control, enabling more reliable grasping and manipulation.

To support training, Xiaomi is also using haptic gloves to capture human interaction data. This approach allows operators to transfer real-world tactile signals directly into training datasets, accelerating learning cycles. It reflects a growing emphasis on data pipelines in robotics, where physical interaction data is becoming as important as simulation.

Durability and Thermal Management Signal Deployment Focus

The company has also addressed one of the less visible but critical barriers to deployment: reliability. Earlier versions of the hand reportedly failed after fewer than 10,000 repetitive operations. The updated design extends this to over 150,000 grasping cycles, a threshold more consistent with industrial use.

In parallel, Xiaomi introduced a bionic “sweat gland” system for active cooling. Using liquid channels embedded within the structure, the system dissipates heat generated during continuous operation. Thermal management is often overlooked in robotics discussions, but it becomes essential when systems move from short demonstrations to sustained workloads.

These improvements build on prior internal testing, where the robot demonstrated multi-hour operation in factory-like conditions with a reported success rate above 90%. While still early, such metrics indicate a transition toward measurable performance benchmarks.

Taken together, the upgrades to CyberOne’s hand highlight a broader shift in humanoid robotics. Progress is no longer defined solely by headline capabilities, but by incremental advances in reliability, sensing, and integration.

For Xiaomi, the development signals a deeper commitment to robotics beyond consumer electronics. For the industry, it reinforces a key reality: achieving human-level manipulation will depend less on singular breakthroughs and more on the convergence of mechanical design, sensory feedback, and scalable data systems.

Humanoid robots are beginning to step out of factories and into the spotlight.

Shenzhen-based LimX Dynamics has unveiled its latest humanoid robot, Luna, at a consumer-facing event in China, marking a notable shift in how robotics companies are positioning their machines – not just as industrial tools, but as interactive systems designed for public environments.

The robot made its debut at the Taobao Influencer Festival, where it performed a catwalk demonstration, showcasing balance, coordination and fluid motion rather than task-specific industrial capabilities.

The choice of venue underscores a broader trend: robotics companies are increasingly targeting applications that involve direct human interaction, from retail and entertainment to hospitality and public services.

From Industrial Platforms to Lifestyle Robots

LimX’s earlier humanoid platform, OLI, was built for industrial use, focusing on durability and operation in demanding environments such as construction sites.

Luna represents a departure from that approach.

With a more refined design and human-like proportions, the robot is intended for environments where appearance, movement and interaction matter as much as functionality. The system features 33 degrees of freedom, enabling more complex motion patterns and smoother walking dynamics.

During its public demonstration, Luna executed a controlled catwalk and an “illusion turn”, a movement designed to test dynamic balance and coordination.

These demonstrations, while not tied to specific industrial tasks, highlight the importance of mobility and expressiveness in robots designed for shared human spaces.

Hardware Meets Human Interaction

Under the surface, Luna builds on the same computing architecture as LimX’s industrial robots.

The system integrates multiple perception technologies, including depth cameras, RGB vision and LiDAR, combined with simultaneous localization and mapping (SLAM) to navigate dynamic environments.

Its software stack is based on ROS 2 and runs on high-performance edge computing hardware, enabling developers to build custom behaviors and applications.

This combination of advanced hardware and flexible software reflects a broader shift in robotics toward platforms that can be adapted for multiple use cases.

Rather than designing robots for a single task, companies are increasingly building general-purpose systems that can operate across different environments with appropriate software layers.

A New Phase for Humanoid Robotics

The unveiling of Luna comes at a time when humanoid robotics is expanding beyond early industrial pilots.

Companies are experimenting with applications that require not only physical capability but also social presence – guiding customers, interacting with audiences or participating in public events.

This shift introduces new challenges.

Robots operating in public spaces must meet higher expectations for safety, reliability and human-like behavior. Movement quality, appearance and interaction design become critical factors in user acceptance.

LimX’s recent funding round suggests that investors are backing this broader vision, betting that humanoid robots will find roles beyond traditional automation.

For now, demonstrations like Luna’s catwalk remain symbolic. But they point to an evolving industry where the success of humanoid robots may depend as much on how they move and interact as on what tasks they perform.