Back to feed
News Story
The AI Insider
1 sources

Singapore Opens Center to Train Humanoid Robots for Police & Public Safety Roles

Singapore's Home Team Science and Technology Agency (HTX) opened the 29,000-square-foot Home Team Humanoid Robotics Centre (H2RC) to develop embodied AI systems for public safety. The center will train and test humanoid robots for tasks such as hazardous-material handling, post-blast investigations, prison-cell searches, and ambulance supply restocking, with the first robot targeted for operational use by 2028. HTX is partnering with ST Engineering on robotics and cybersecurity and with Singtel RE:AI to expand sovereign AI infrastructure using Nvidia Grace Blackwell GB300 GPUs.

SynthePulse Insight · AI deep reading

Singapore Sends Embodied Intelligence Into the Home Team: H2RC's 2028 Timeline and Data Threshold

Version 1 · 1 source

Singapore's Home Team Science and Technology Agency has opened the 29,000-square-foot H2RC, with the goal of putting its first humanoid robot into service before 2028. What really determines success is not the robot body itself, but public-safety-specific data, computing infrastructure, and pre-deployment cybersecurity and reliability validation.

  • HTX has opened H2RC, focused on embodied-intelligence humanoid robots for public safety, with a center area of 29,000 square feet.
  • The goal is to put the first humanoid robot into service before 2028, serving Home Team agencies including the police, civil defence, immigration and prisons.
  • Application scenarios include handling hazardous materials, investigating post-explosion sites, searching prison cells for contraband, and restocking ambulances with emergency medical supplies.
  • HTX signed a memorandum of understanding with ST Engineering and is working with Singtel RE:AI to upgrade the NGINE sovereign AI infrastructure with Nvidia Grace Blackwell GB300 GPUs.
  • HTX says public safety robots need specialized data that cannot be bought or borrowed directly from industrial applications, and factory training data cannot automatically transfer to emergency scenes.
  • Cybersecurity and reliability testing must be completed before deployment into real operations, and development must go through capability building, trials and validation stages.
Open section navigationOne Center, One 2028 Timeline

One Center, One 2028 Timeline

Singapore's Home Team Science and Technology Agency (HTX) has opened the Home Team Humanoid Robotics Centre (H2RC), focused on developing embodied AI systems for public safety, and plans to put its first humanoid robot into service before 2028. The center covers 29,000 square feet and will train and test robots, serving agencies across Singapore's Home Team system including the police, civil defence, immigration and prisons.

This positioning sets H2RC apart from most commercial humanoid robot development. HTX says most commercial humanoid robot development is concentrated in manufacturing, household and consumer uses, with less development for public safety applications. Singapore's Minister for Law and Second Minister for Home Affairs Edwin Tong said at the opening ceremony that H2RC is the first such center dedicated to public safety humanoid robots — this "first" statement is the minister's own claim and has not been verified by a third party.

Why Humanoid: The Constraints Behind the Form Choice

HTX says it chose the humanoid form because its human-like shape allows it to use existing tools in workspaces already designed for people. This rationale points to a real constraint: the robot needs to operate without modifying existing workspaces and tools.

Application scenarios HTX is considering include handling hazardous materials, investigating post-explosion sites, searching prison cells for contraband, and restocking ambulances with emergency medical supplies. Edwin Tong said the Home Team views this technology as a means to augment rather than replace officers, especially given that Singapore faces manpower constraints and operates in a dense urban environment.

The Real Threshold: Public-Safety-Specific Data

H2RC will support the full embodied AI development cycle, including motion capture, teleoperation, human demonstrations and simulation used to generate training data. The center will also train and evaluate vision-language-action (VLA) models, connecting what the robot sees and understands with the physical actions it takes.

HTX points out that public safety robots need specialized data that cannot simply be bought or borrowed from industrial applications, and H2RC will use realistic models and operational scenarios to collect, review and annotate data. Edwin Tong said robots trained on the factory floor will not automatically know how to operate safely at an emergency scene, so Home Team-specific data is at the core of the program.

This means H2RC's output is not just robot hardware, but a data collection and annotation process. Whether the center's 29,000-square-foot scale matches the amount of training data required has not been disclosed.

Computing Power and Industry Partnerships: Who Is Supporting This System

HTX and ST Engineering signed a memorandum of understanding at the opening ceremony, covering joint research and development, technology and platform integration, industry and academic collaboration, and cybersecurity. HTX is also working with Singtel RE:AI to upgrade its NGINE sovereign AI infrastructure with Nvidia Grace Blackwell GB300 GPUs, to support embodied AI and multimodal model training for H2RC and other Home Team applications.

These two partnerships correspond respectively to the robot body and integration, and to the computing foundation. However, the specific outputs and timelines of the partnerships with ST Engineering and Singtel RE:AI have not been explained.

Hard Constraints Before Deployment: Cybersecurity and Reliability

Development will go through capability building, trials and validation stages, and cybersecurity and reliability testing must be completed before deployment into real operations. This is a clear pre-deployment condition, and it also means the 2028 goal is not simply a technology demonstration deadline.

The actual performance and reliability of VLA models in public safety scenarios have not been disclosed, and the real-world effectiveness of humanoid robots in scenarios such as hazardous materials and post-explosion sites has not been verified. Whether the first humanoid robot can enter service on schedule before 2028 remains uncertain.

Credibility boundary

This article is based on a report by The AI Insider (article_id 19788). HTX's opening of H2RC, the 2028 goal, the 29,000-square-foot area, the agencies served, the application scenarios, the development cycle, VLA model training, the ST Engineering memorandum of understanding, the Singtel RE:AI and Nvidia Grace Blackwell GB300 GPU partnership, and the pre-deployment cybersecurity and reliability testing requirements all come from the report's confirmatory statements. Statements such as the rationale for choosing the humanoid form, that public safety application development is less common, the "first such center" claim, augmenting rather than replacing officers, that public safety requires specialized data, and that factory training data cannot transfer directly are source_claims from HTX or Edwin Tong, and this article does not rewrite them as established facts. Whether the 2028 target can be met on schedule, the actual performance of VLA models, the specific outputs and timelines of the partnerships, the match between the center's scale and data volume, and real-world effectiveness are all uncertainties that have not been disclosed or verified.

Insight takeaway

The core of H2RC is not building a humanoid robot, but establishing a dedicated data, computing and validation process for public safety. Whether the 2028 goal can be delivered depends on whether this process can run through under the hard constraints of cybersecurity and reliability testing.

Primary report

The AI Insider

Primary source