Humanoid robots are still maturing — experts urge a crawl-walk-run approach

Full text

It’s no longer a question of if – only when. The promise of humanoid robotics is finally crystallizing into a commercial reality.

This new era will create a sector with the heft to rival megaliths like automotive and computing.

But we’re not there yet, and though large and well-known manufacturers are investing billions in development, with some clear leaders, the robots themselves remain in adolescent stage.

My advice? Treat them like any other adolescents and don’t rush them into adulthood before they’re ready. Take a crawl walk run approach to innovation, build capability layer by layer, and let maturity emerge from momentum. This is the fastest way to win the race to commercial deployment and value.

Development teams are making progress by the day, stimulated and inspired by the obvious appeal of a robot shaped like a human, operating deftly and effectively in any environment designed for people, using existing tools and infrastructure . But we’re clear on the obstacles, and equally clear with business leaders on the need to understand the technical, practical, regulatory and social factors that stand between them and the market.

That said, there’s no doubt in my mind that humanoid robotics represents not just a new way to win – but a way to win big. Companies that are first to crack the challenges I’m about to describe will seize powerful advantages. They’ll define industry standards, accumulate proprietary data and build customer relationships that late entrants can’t hope to replicate.

Early deployment – even for tasks like packing boxes – generates real-world learning that accelerates improvement.

The human-shaped challenge

The formidable gap between impressive (if carefully pre-choreographed) demonstration videos and actual deployment is marked with several key technical challenges, not least the physics of the human form. Balance and locomotion are among the hardest problems in robotics. We walk with a complex, energy-efficient gait that takes years to learn. Replicating for a robot requires real-time processing of sensor data, continuous adjustment to shifting weight and the ability to recover from unexpected disturbances.

Reinforcement learning and improved actuators have produced robots that can walk, run, and even perform parkour in controlled settings. But real-world environments are a chaotic mess of uneven floors, unexpected obstacles and slippery surfaces – and current systems still struggle here.

Dexterity and manipulation are equally daunting. Human hands have 27 degrees of freedom and extraordinary tactile sensitivity. Once learned, we can thread a needle, crack an egg, or catch a ball without conscious thought. Robotic hands have improved substantially, but fine motor control, delicate force application and adaptive grip remain limited. Tasks that seem trivial to us are extraordinarily difficult for machines.

Our research notes the promise of perfecting fine manipulation as physical AI teams progresses from lab proof-of-concept to stable pick-and-place cycles with real hardware. It’s tough, because it’s all about building new capabilities from scratch. But with a growing confidence in areas like fine manipulation, human-robot interaction and whole-body control, we’re moving ever closer to significant breakthroughs.

A new way to look at perception

Perception and decision-making represent further technical hurdles. Robots must interpret cluttered, dynamic environments in real time, distinguishing between a crumpled napkin and a spilled hazard, recognizing when a human is about to cross their path, and making split-second decisions about how to respond. Current AI can handle many of these tasks in isolation, but integrating them into a coherent, reliable whole is a work in progress.

As these technical problems are solved, economic viability will come increasingly into focus. While some humanoids are advertised with a cost of a few thousand dollars, these are essentially expensive toys rather than effective workers. Top-of-the-range humanoid robots cost hundreds of thousands of dollars, far more than what most businesses can justify for tasks that humans perform adequately.

Manufacturing at scale could bring prices down, but the path to a $20,000 or $30,000 unit that could potentially result in productivity and cost efficiency gains remains uncertain. Robot-as-a-Service financing models will enable early up-take but the underlying cost challenge will remain a blocker to mass adoption.

Getting closer to real-world usefulness

Personally, I’m optimistic about overcoming the remaining challenges. There is work to be done on operational reliability (we’re not yet there with machines that operate autonomously for extended periods without much intervention); legal frameworks; physical safety (regulatory guidelines are in their infancy); and even public perception (resistance to automation has derailed past initiatives.)

The key is to stay focused on the end game. With the remaining hurdles diminishing, the direction of travel is irresistible. Every small breakthrough – like that box packing example I mentioned earlier – brings us closer to real-world usefulness and every early deployment teaches us something we can’t learn in the lab. Build steadily, crawl-walk-run, and the rewards await the ambitious first movers.

70+ of the best AI tools rated and ranked .

This article was produced as part of TechRadar Pro Perspectives , our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit