MORPHI unveils MoRA brain and KINO home robot at WRC 2026: 15 minutes of continuous housework

MORPHI Intelligence (墨奇智能), a general-purpose embodied AI company founded only half a year ago, delivered its first public exam at WRC 2026: the debut of MoRA (MORPHI Reasoning & Autonomy), its self-developed embodied model architecture, together with MORPHI KINO, its wheeled robot built for home-service scenarios. It is also the company's first appearance at a major exhibition.

The showstopper was a 15-minute live household task. KINO cleared trash from a coffee table, walked to a fridge, checked its stock, fetched a bottle of water from another table to restock it, moved wet clothes into a dryer, and then folded and stacked the dried laundry — all without any human commands. Staff said the robot was given only simple goal prompts such as “clean this room, tidy the table, put the water in the fridge, and handle the laundry”; the rest was orchestrated internally by MoRA.

The company is co-founded by Huang Qingqiu (黄青虬), a former head of Huawei's autonomous-driving AI team and a “Huawei Genius Youth” awardee, who argues that world models and VLA are not the answer. MORPHI's bet is “Agentic-Native” embodied intelligence: instead of bolting an action policy onto a large model, it embeds goal persistence, execution memory, and progress tracking natively inside the policy model's System 1.

MORPHI contends the conventional hierarchical approach — a System 2 language/vision model decomposing goals into fixed-format text sub-task commands for a System 1 policy — suffers from limited goal interfaces and short, reactive execution; every translation layer loses information, and any environmental disturbance forces the robot to stop and wait for re-planning. MoRA's System 1 carries three levels of memory — short-term control continuity, mid-level step state, and long-term task goals — letting the robot judge progress while outputting actions, sense and correct most mid-course deviations by itself, and only request System 2 intervention at its capability boundary.

Beyond the main demo, the booth featured pick-and-place organization, a mystery-box delivery interaction, and photo sessions: visitors could freely interfere with the picking process to test real-time closed-loop adjustment, and could decorate the robot's face with magnetic expressions before taking photos while it posed.

Data is the other pillar of MORPHI's strategy. The company argues that for embodied AI, real-scene data quality matters far more than quantity, and it follows an optimized “no-bot” collection route: starting from human first-person-view data, aligning human behavior with robot actions in a unified action space, and capturing real work with its self-built MORPHI Sense Kit, which claims millimeter-level trajectory reconstruction even in low-texture, high-reflection scenes. Collection devices have already entered hotels and serviced apartments, where actual workers generate data during daily shifts.

MORPHI says it has accumulated 30,000 hours of real-scene data and plans to reach 150,000 to 200,000 hours by the end of 2026. Asked whether entering the embodied race three and a half years late is too slow, the company's answer is that embodied intelligence is a marathon.

In 2026 the field has returned to scenario orientation, and long-horizon tasks are becoming the new theme. MORPHI is choosing not to chase world-model hype or paste on the VLA label — instead it is making long-horizon capability native to the model and steadily building data infrastructure in real scenes. This startup, valued at over 7 billion yuan, is choosing to accelerate in the middle of the marathon.