Physical AI startup Real World (RLWRLD) will demonstrate its robotic technology in Tokyo on October 15 and 16, 2026, through a collaboration with Mitsubishi Estate and the Tokyo Marunouchi Innovation Platform (TMIP). The trial will take place along Otemachi Nakadori, where robots powered by RLWRLD’s RLDX-1 model will hand beverages to visitors and collect empty plastic bottles. The demonstration forms part of TMIP’s citywide testing initiative and Wellness Weeks 2026 Autumn, allowing the partners to examine how AI-controlled machines can deliver services and interact with people in public spaces.
Testing Robots in Everyday Interactions
During the first activity, a robot will grasp a beverage and pass it directly to an office worker or visitor. A second activity will involve picking up empty containers that participants have placed in designated positions on tables. These actions will provide opportunities to observe robotic handling and human interaction outside a controlled laboratory.
The partners see potential applications in supplying materials, collecting items, and supporting workers in industrial and service environments. They also want to assess whether robots can operate safely and consistently as people and surroundings change. The organizers caution that activities may be modified after preliminary tests or canceled because of weather or equipment conditions.
Technology Behind RLDX-1
The demonstration will use robots powered by RLWRLD’s RLDX-1 robotics foundation model, which is designed to work with five-fingered robotic hands. The system combines information from vision, touch, and movement to support tasks requiring precise physical manipulation. RLWRLD develops its models using multimodal data captured from industrial settings, including manufacturing and logistics.
The company introduced RLDX-1 earlier in 2026 and has reported strong results in simulation-based benchmark evaluations against other robotics models. According to its own testing, RLDX-1 scored 70.6 on RoboCasa Kitchen and 58.7 on GR-1 Tabletop. Those reported scores are not evidence that the technology has already achieved the same performance in public urban environments, which makes practical trials important.
Collaboration Across the Marunouchi District
RLWRLD operates from EGG, a startup hub in Tokyo’s Shin-Marunouchi Building, and has reported seed backing from Japanese and South Korean manufacturers and venture investors. Mitsubishi Estate is collaborating with the startup through TMIP to explore potential applications of physical AI in the Marunouchi area. The demonstration will provide an opportunity to examine how advanced robotic systems could support services and operations in urban environments.
The initiative operates under TMIP’s program for collaborative testing across Otemachi, Marunouchi, and Yurakucho. TMIP brings together corporations, startups, public institutions, and academic organizations to develop new products and address shared challenges. Its network includes more than 400 member and partner organizations, according to the platform.
Beyond the Public Demonstration
Japan’s labor shortages are one reason companies are exploring robots capable of performing more varied physical tasks. However, adapting sophisticated AI models to real workplaces requires more than promising laboratory or simulation results. Developers must also establish operating procedures, reliability, and acceptable ways for machines to interact with people.
The partners intend to use lessons from the trial to evaluate future applications of physical AI across industries. The event will also give visitors a direct encounter with robotics technology that is usually developed in specialized facilities. No wider deployment timetable or commercial rollout has been announced as part of this demonstration.
RLWRLD’s upcoming Tokyo trial will test how AI-enabled robots handle everyday objects and interact with the public in a busy business district. Through Mitsubishi Estate and TMIP, the startup will gain access to a real-world setting for examining practical performance and public acceptance. The findings could inform future uses of physical AI, although the two-day experiment alone will not establish readiness for large-scale deployment.