TRENDING that moves from niche to universal adoption in just a few decades.
What’ s behind this change? Ageing workforces, rising wages and legacy infrastructure that rigid machines struggle to navigate. Humanoid robots, trained through simulation and powered by AI, are built to function in spaces designed for people.
Legacy factories, modern problems
Labour shortages are now a long-term problem. In Japan, over 28 % of the population is over 65. In Europe
Robots are stepping into human roles without forcing a redesign of the entire facility.
Zeeshan Mehdi, Engineering Director –
Middle East at SoftServe and North America, ageing populations are growing fast. With declining birth rates and fewer young workers entering industrial jobs, relying on labour import is no longer a sustainable option.
Most factories and warehouses weren’ t built for robots. Their workflows depend on human-like tasks: climbing stairs, opening doors and using tools. These are things fixed automation can’ t do well. Traditional automation was built for simple, repeatable tasks – not dynamic, human-like environments. It is costly to integrate with legacy equipment built for human use, cannot adapt to new workflows without reprogramming and lacks the mobility and dexterity compared to humans. To stay competitive, you need automation that works where people do – and adapts as quickly as the job changes.
Bridge the Sim2Real gap
Simulation-first training significantly speeds up development and de-risks early-stage deployment – but it isn’ t the whole story. A well-known challenge in robotics is the Sim2Real gap: the difference between how a robot performs in simulation and behaves in the real world.
Developing humanoid robots combines synthetic data, domain randomisation and real-world fine-tuning to bridge this gap. Simulation helps shape foundational skills, but hands-on testing is essential. No amount of virtual training fully replaces physical-world adaptation.
Simulation-first robotics starts in the virtual gym
The latest humanoid robots aren’ t just manually programmed – they can be pretrained using Machine Learning techniques. Their behaviour is shaped in photorealistic, physics-driven virtual environments before deployment.
Although simulation accelerates development, a Sim2Real gap remains – making physical testing and fine-tuning essential.
Simulation workflows allow humanoids to learn motor behaviours, balance and manipulation policies. They also support scalable training of control loops across multiple robots, enable testing of failure recovery scenarios without risk and provide high-fidelity transfer to real robots.
NVIDIA Isaac GR00T: the future of intelligent humanoid robotics
NVIDIA Isaac GR00T represents a new class of robotics architecture – a general-purpose foundation model built for embodied AI. Trained on a blend of web-scale data and synthetic motion sequences, GR00T is designed to unify vision, language, planning and control.
GR00T integrates with NVIDIA’ s threecomputer architecture to support simulation-todeployment workflows: NVIDIA DGX trains the foundation model, NVIDIA OVX( RTX GPUs) powers synthetic environment generation and Jetson Thor runs onboard inference for real-time autonomy at the edge.
As foundation models like GR00T evolve, they hold the potential to accelerate multi-task autonomy, reduce engineering overhead and expand deployment options.
52 INTELLIGENTCIO MIDDLE EAST www. intelligentcio. com