How humanoid robots are likely to take over odd, hard jobs

Late in the evening, a busy apartment tower in Bengaluru is awake. The lift doors open; Out come the delivery bags, small boxes, and tired commuters. A young helper steps out first. Next to him walks a two-legged robot with a soft cover. Their job is simple. They carry water cans from the lobby to the service corridor. They restock a few groceries in the small store. They push the garbage cart to the sorting area. The child watches and waves. The robot stops, tilts its head like a smile, and waits. When the child signals, it moves again. Nothing dramatic. Just normal life. This is how the next decade of robotics may feel: familiar places running a little easier.

It is dawn at a shipyard on the west coast. Birds circle; welders greet each other; a humanoid robot in a hard hat walks through a gap in a half-built ship. There is no big moment. Only quiet work. The robot feeds a cable through a metal wall. It carries a sensor into a tight space. It checks if paint covers the area behind an awkward rib. By noon, a clean progress report shows up on the foreman’s table. There is a gentle reminder to plan a battery top-up. It is a small scene, but it says a lot. Robots made for human places can do the jobs we hate doing. They do them with care and attention.

And they still have limits. Across the United States, the actors behind these moments are suddenly star-studded. Agility Robotics’ Digit is a quiet workhorse for back-room tote moves; Apptronik’s Apollo targets light-industrial tasks; Sanctuary AI and Norway’s 1X focus on dexterous, hand-led work; Tesla’s Optimus is training on Tesla’s own lines; Boston Dynamics’ Atlas keeps stretching what legs and joints can do; and Figure AI’s new Figure 03 adds a softer, home-safe posture with tactile-rich hands and quick ‘sip-charging’ between tasks. China, meanwhile, has turned humanoids into a national project: Unitree iterates at speed (H1/G1/R1), Fourier Intelligence opens developer-friendly platforms, UBTech courts service roles, and vocational school training lines at firms like Leju/Kuavo churn out the messy data robots need to handle the real world.

India and Europe bring their own signatures. From India, Bengaluru’s Ati Motors blends proven mobile autonomy with humanoid-style manipulation—a pragmatic “do the job that pays for itself” approach for shop floors and back rooms. From Europe, Barcelona-based PAL Robotics (REEM-C, TALOS) anchors a decade of rigorous biped control and human-robot interaction research. Put together, the ecosystem feels like a relay: the US pushes speed, China pushes scale, Europe pushes discipline, and India pushes practicality.

Still, this is an evolving story. Three stubborn truths can explain why.

Reliability faces everyday chaos. Real sites are full of puddles, torn labels, reflective packaging, cable snags, and ramp lips that look unimportant until they aren’t. In the apartment tower, a water can sweat and slip; the humanoid needs to re-grasp without crushing it or blocking the lift. Success here isn’t a viral backflip; it’s graceful degradation—finishing the job more slowly and safely when the world misbehaves. That’s what converts pilots into purchase orders.

Total cost is more than its capability - cost per successful task measured over thousands of cycles with predictable variance. In adjacent service-robot categories (mostly wheeled today), outcome-priced deployments show useful bookends: productivity lifts near the 50–60 per cent range and 15–25 per cent labour offset when workflows are stable and operations are tuned. Humanoids will need to beat those numbers to graduate from fun demos to real business.

Machine learning needs hoards of data. Competence only arrives one edge case at a time. China’s answer to this is open testbeds and shared stacks that accelerate the loop; the US advantage is deep mechatronics paired with embodied-AI talent. The teams that capture, share, and reuse data best will generalise fastest.

And then there are batteries: Legs and hands gulp energy. Stop-start motion, high-torque grips, and constant balance corrections drain batteries very quickly. The chemistry of battery packs can deliver strong bursts; the tougher milestone is hours of safe, solid work - especially in heat and humidity -without burning up components. Robot manufacturers will use this tactic for now: swappable packs, short wireless top-ups between tasks, and improved thermal paths.

Where the market is and where it could be. The rails that humanoids will initially ride are already being laid by professional service robots in logistics, cleaning, and hospitality. Shipments for these categories in 2024 sat near the 200,000-unit mark, up roughly high single digits year-on-year. India’s broader robot installations (all types) are rising fast from a low base; China’s manufacturing robot density has climbed into the global top tier. Looking ahead, models project humanoids industry as a multi-tens-of-billions category by the mid-2030s. A cautious midpoint potential can be around US$35–40 billion by 2034–2035, with annual shipments crossing 1 million units in optimistic scenarios - assumptions that hinge on reliability, learning efficiency, and falling actuator/battery costs. The direction of travel is clear; the slope depends on engineering. Approx price per unit was about Rs. 2 crore in 2023, and Tesla is gunning for about Rs 15-20 lakhs per unit when they get their Optimus into the market.

So how should businesses engage - without overcommitting? The emerging consensus is to experiment deliberately. Define a narrow, repeatable slice of work; write service-level expectations that matter and track unit economics like a hawk. Robotics-as-a-Service (RaaS) - subscribing by month or by outcome instead of buying outright is looking like a sensible way to try that. It is not a silver bullet. But for festival peaks in retail, night logistics in hospitals, or short bursts of inspection in heavy industry, capability to match business needs, and let teams “earn trust in production” before scaling.

Back at the shipyard, the humanoid reverses out of the hull, hands off its notes, and steps aside as a welder squeezes past. Over in the apartment tower, the helper finishes its loop and returns to a small dock near the security desk. The guard signs off on the checklist, glances at the live camera feed, and smiles as the late-night elevator crowd thins. Two very different places, one quiet thread: human spaces becoming slightly easier to run.

If there’s inspiration to take from this moment, it’s a simple, human one. The real promise of humanoids isn’t to replace the pride we take in our work; it’s to retire the parts of work that hurt our backs, fray tempers, and steal time. The US–China sprint will keep compressing development cycles; Europe’s discipline and India’s practicality will keep deployments grounded; batteries and edge cases will keep engineers humble. And progress will look less like a single break through moment and more like a steady stream of small wins that add up - fewer dropped parcels on a rainy night, fewer near-misses in tight corners, fewer hours lost to drudgery. If we start small, measure hard, and scale only when the numbers say so, then by 2034 the headline won’t be about robots arriving. It will be about teams—human and humanoid—delivering steadier output, safer sites, and better jobs, the kind where skill and judgment matter more because the boring grind has someplace else to go. That’s not science fiction; it’s progress you can feel between a foreman’s tea break and a family’s late-evening lift ride home.

(The author is the Chairman & CEO of Brightcom Group)