The Next 10 Years of Robotics: A Grounded Forecast to 2036

Robotics forecasts have a worse track record than almost any field in tech. We have been ten years away from the home robot for forty years. The reason is a recurring mistake: assuming that because a robot can do something impressively in a demo, it can do it reliably, cheaply, and safely in the real world. Those are different problems separated by years and billions of dollars.

The deeper reason has a name: Moravec's paradox (Hans Moravec, Mind Children, 1988). The things that feel hard to us (chess, integrals, legal reasoning) are cheap to automate, because they are recent, shallow, and symbolic. The things a two-year-old does without thinking (grasp a spoon, cross a cluttered room, recover from a slip) are a billion years of evolutionary optimization and staggeringly hard to reproduce. Software AI ate the top of that stack; robotics is stuck at the bottom.

This forecast tries to respect that gap. It extrapolates the real trends, foundation models reaching into the physical world, actuator and compute costs falling, simulation improving, while naming the bottlenecks (data, contact physics, reliability) that keep bending the optimistic curves. The rule throughout: a capability is real when its sustained failure rate clears the bar for the job.

Key predictions

• Foundation models eat robotics' software stack. The hand-tuned, task-specific pipelines give way to learned, general policies: vision-language-action models trained on broad data. This is the decade's biggest shift, and it's already underway.
• The data bottleneck is the whole game. There is no internet-scale dataset of robot actions. Whoever solves data collection (teleoperation, simulation, or learning from video) wins. This is the binding constraint.
• Humanoids stay overpromised. They make spectacular demos and real progress in factories, but a reliable, affordable, general-purpose humanoid doing your housework is not a 2030s consumer reality. Bet on constrained commercial deployment.
• The visible action moves to warehouses, logistics, and manufacturing: structured environments where the economics already work and reliability is achievable.
• Hardware quietly gets cheaper and better: actuators, sensors, and on-robot compute follow their own cost curves, making capable robots economically viable in more places each year.
• Sim-to-real keeps closing but never fully closes. Simulation does more of the training; the real world keeps its veto.
• The bottleneck shifts from "can it move?" to "can we trust it?": safety, reliability, and certification become the hard problems, exactly as they did for industrial automation.

Predictions at a glance

How to read a robotics forecast

Two rules keep it honest. First, separate the demo from the deployment: a robot folding laundry on YouTube is years from a robot that folds laundry in ten thousand homes without breaking, hurting someone, or needing a babysitter. Second, follow the data and the dollars: a capability ships when someone can collect enough data to train it and the unit economics beat the human or the fixed automation it replaces. Everything else is a tech demo.

The near term: 2026-2028

VLA models move from research to the floor (high confidence). The single biggest change in robotics is happening in software. The task-specific pipelines are being rewritten. Instead of hand-coded motion planning and bespoke perception per task, robots increasingly run learned policies that map what they see and what they're told straight to action. Expect this to work first in narrow, high-value tasks (bin picking, machine tending) and stay brittle at the edges. The companion shift: roboticists now use AI models like Claude to write control code, generate simulation scenarios, and label data. AI is eating the development of robots as much as their behavior.

Teleoperation becomes a data strategy on top of a control mode (high confidence). Because robot-action data is the bottleneck, human teleoperation gets repurposed as the way to collect training data at scale. The companies that build the best data flywheels pull ahead.

Put numbers on it and the wall is obvious. A language model trains on ~10^13 tokens scraped for free. The largest open robot-manipulation corpus, Open X-Embodiment (Google DeepMind and ~30 labs, 2023), is on the order of 10^6 trajectories, each physically executed by a real or teleoperated arm in real time. There is no crawler for the physical world. If scaling laws hold in the embodied regime, loss falling as a power law, L(D) ≈ L∞ + (D₀/D)^α with α well under 1, then halving a policy's error demands multiplying the data, and every trajectory costs seconds of robot time plus real wear. Free tokens versus expensive action tokens: that asymmetry is why robotics will not simply inherit the LLM curve, and why whoever drives the marginal cost of one labeled trajectory toward zero wins. Data logistics is the moat.

Warehouses and logistics keep scaling (high confidence). This is where robotics already pays for itself. AMRs, picking, and sortation expand because the environment is structured and the ROI is clear. Boring, real, and where the money is. Amazon crossed one million deployed warehouse robots in mid-2025 (per Amazon and reported by CNBC and the Wall Street Journal), against roughly 1.56 million human employees. That is the shape of the near-term future: fixed and mobile automation in structured buildings, at a scale no humanoid program is close to matching. See warehouse and logistics robotics for how those fleets are actually built.

The data problem has three exits

If the binding constraint is robot-action data, there are exactly three ways to manufacture it, and every serious lab is placing bets across all three. Each buys data at a different price and a different quality.

1. Teleoperation (expensive, high fidelity). A human drives the robot; the robot logs its own states and actions as ground-truth demonstrations. Stanford's ALOHA and Mobile ALOHA showed you can collect competent bimanual demonstrations on a sub-$32k rig. The DROID dataset (2024) pushed the open frontier to 76,000 trajectories across 564 scenes and 86 tasks on Franka arms, and Open X-Embodiment (Google DeepMind with ~22 institutions, 2023) pooled on the order of one million trajectories across 20-plus robot embodiments. Every one of those trajectories cost real robot-seconds and a human operator's attention. Teleoperation is the gold standard for label quality and the worst option for marginal cost. See robot teleoperation for the control-mode mechanics underneath.

2. Simulation (cheap, reality-gapped). Spin up thousands of parallel worlds in NVIDIA Isaac Sim or Isaac Lab, randomize physics and appearance, and harvest millions of episodes overnight for the price of GPU time. NVIDIA's GR00T N1 (March 2025) leaned on exactly this, mixing Omniverse and Cosmos synthetic data with real captures. The catch is the one named in this forecast's "what will not happen" section: you can only randomize what you can model, and contact, friction, and deformation are the parts you model worst. Simulation is nearly free per episode and systematically wrong in the regime where manipulation robots earn their money. See sim-to-real transfer for how teams try to close that gap.

3. Human video (nearly free, hard to use). The internet has billions of hours of people doing manual tasks. The problem is that video has no action labels and the wrong morphology: a hand is not a two-finger gripper, and there is no recorded joint torque behind a YouTube clip. Projects like the Universal Manipulation Interface (UMI), a handheld gripper that records human demonstrations in the robot's own action space, exist precisely to bridge that gap. Human video is the cheapest data on Earth and the hardest to convert into something a policy can execute.

Rule of thumb: teleoperation buys quality, simulation buys volume, human video buys reach. The winning data strategy blends all three and drives the marginal cost of one usable labeled trajectory toward zero. Whoever does that first sets the pace of the decade.

The VLA model race

The software shift has a concrete leaderboard now, and it is worth naming the players because the architecture is converging fast. See the foundation models and VLA flagship for the full technical treatment; this is the map.

• RT-2 (Google DeepMind, 2023) was the proof of concept: co-fine-tune a vision-language model on robot trajectories and web data, and web-scale semantic knowledge transfers into action. It could pick "the extinct animal" from a table of toys. Brittle, slow, and a genuine turning point.
• π0 / openpi (Physical Intelligence) is a flow-matching VLA pretrained on more than 10,000 hours across seven robot configurations and 68 tasks, augmented with Open X-Embodiment and DROID. Physical Intelligence open-sourced the weights, which matters: it puts a capable generalist policy in reach of teams that cannot fund their own data flywheel.
• GR00T N1 (NVIDIA, March 2025) is a 2-billion-parameter open foundation model for humanoids with a dual-system design, trained on a blend of Omniverse/Cosmos synthetic data and real humanoid captures.
• Helix (Figure, February 2025) made the split explicit: System 2 is an internet-pretrained VLM running at 7 to 9 Hz for scene understanding and goal sequencing, and System 1 is a visuomotor policy running at 200 Hz that drives all 35 degrees of freedom.
• Gemini Robotics (Google DeepMind, 2025) brings the Gemini model family into the loop, betting that a frontier multimodal model plus an action head generalizes better than a purpose-built small policy.

The pattern underneath all of them is a two-system split that echoes Kahneman: a slow, semantic "think" model reasoning at single-digit hertz, and a fast, reactive "act" policy closing the loop at hundreds of hertz. The slow half inherits web knowledge and reasons about novel instructions. The fast half handles contact and balance where latency is the enemy. Fusing the two cleanly, without the fast loop starving or the slow loop lagging, is where much of the near-term engineering goes, and it is why edge compute on the robot is back on the critical path.

The mid term: 2028-2032

General-purpose manipulation gets good enough in constrained settings (medium confidence). A robot arm that can be told, in plain language, to do a new pick-and-place task and just do it, reliably enough for a factory, becomes real. General manipulation in unstructured homes stays hard.

Humanoids find their actual niche (medium confidence). After the hype cycle, humanoids settle into roles where a human-shaped body genuinely helps in human-built environments, some warehouse and manufacturing work, while most automation continues to use the right shape for the job (arms, AMRs, gantries), which is rarely humanoid. The form factor is a marketing magnet and an engineering tax.

Quantify that tax and the caution writes itself. A bolted-down 6-DOF arm is well-conditioned: fixed base, known workspace, gravity compensated open-loop. A bipedal humanoid carries ~25 to 40 actuated DOF, and its default state is falling. Standing is active stabilization of an inverted pendulum: linearize it and you get θ̈ ≈ (g/L)·θ, an unstable pole at +sqrt(g/L) that feedback must catch every cycle or the machine goes down; locomotion then hangs on keeping the ground-reaction force inside the support polygon (the ZMP criterion, Vukobratović, 1972). On top of that, legs pay a cost of transport, COT = P / (m·g·v), several times worse than wheels on the flat floors warehouses already provide. You take on all that overhead for one thing: operating in spaces built for human bodies. Where that payoff is real (stairs, ladders, mixed human workspaces), humanoids earn their keep; where the floor is flat, the arithmetic sends the buyer back to an AMR.

Hardware cost curves compound (high confidence). Actuators, sensors, and edge compute keep getting cheaper and denser, pulling capable robots into mid-market manufacturing and new sectors (construction, agriculture, inspection) that couldn't justify them before.

Reliability and calibration become the moat (high confidence). As the software gets smart, the differentiator becomes whether it works on the thousandth try as well as the first: the unglamorous world of calibration, real-time control, and safety certification. Demos are cheap; dependability is expensive.

Here is where teams building on learned policies get burned. A 95%-reliable demo is a triumph in a paper and a catastrophe on a line: at a 5% per-cycle failure rate, an unbroken run of just 200 cycles has probability 0.95^200 ≈ 3.5×10^-5, effectively never. The jump from "barely one nine" to the parts-per-million world of industrial automation takes a different kind of engineering discipline, one that more training data alone does not deliver. And certifiers do not accept "the neural net usually works." Industrial arms answer to ISO 10218; power-and-force-limited collaboration to ISO/TS 15066 (biomechanical force and pressure limits for human contact); personal-care robots to ISO 13482; safety-function integrity is argued in IEC 61508 SIL levels. A stochastic, hard-to-interpret VLA policy is genuinely awkward to fit inside frameworks built to demand deterministic, verifiable behavior, and closing that gap is the decade's real long-term work. Read functional safety for robots for how that argument actually gets made to an auditor.

The humanoid scoreboard

The humanoid field is where hype and reality diverge most, so it is worth grading the leaders on evidence rather than reveal videos. The signal to track is sustained work in a real building: totes moved, hours run, interventions per shift. Everything below is company-reported or press-reported unless a peer benchmark exists, and it should be read that way.

Two things fall out of this table. First, the only humanoid with a defensible throughput number in mid-2026 is Digit, and it earns that number in a warehouse, the most structured environment on the list. Second, the pricing spread is telling: a $13.5k Unitree G1 and a humanoid doing certified 24/7 industrial work are not the same product category, and conflating them is how the hype cycle inflates. A cheap dev platform sells because it is cheap; a production humanoid has to clear reliability and safety bars that no consumer-priced machine is near.

The take: grade humanoids on totes-per-hour and interventions-per-shift at a real site, never on a montage. By that standard the leaders are doing genuine, narrow, structured-environment work, and the general home helper is still absent. For a buyer's-eye view, see how to choose a humanoid robot and the humanoid hardware teardown.

The long term: 2032-2036

Plausible: robots are common in commercial and industrial settings and starting to appear in semi-structured public ones (cleaning, delivery, inspection). Foundation models make deploying a robot to a new task a matter of data and fine-tuning rather than months of integration. The field looks less like bespoke engineering and more like the AI software stack.

Genuinely uncertain: whether a truly general home robot becomes affordable and reliable within the decade (probably not), whether one "robotics foundation model" generalizes across bodies and tasks the way LLMs generalize across text, and whether the humanoid bet pays off or becomes the decade's most expensive distraction. Treat confident claims on these with suspicion.

Where robots ship first: the structure ladder

The single best predictor of when a robot deploys is how structured its environment is and how forgiving its failure mode. Rank the jobs by those two axes and the rollout order falls out almost mechanically. This ladder is the practical form of the whole forecast.

Money and reliability climb the ladder from the bottom. A capability lands on a rung when someone can collect enough data to train it and the unit economics beat the incumbent. The home sits at the top because it fails every test at once: no structure, no isolation, no dataset, and a bystander who can be hurt. That is why the honest forecast puts consumer home robots last, and why the warehouse and logistics and industrial arm rungs keep absorbing the capital.

Leading indicators to watch

Forecasts age badly, so here are the measurable signals that will tell you whether the optimistic or pessimistic curve is winning, well before the press notices.

• Cost per usable labeled trajectory. If teleoperation and human-video pipelines drive the marginal cost of one good demonstration down an order of magnitude, the data wall cracks and the timelines pull in.
• Synthetic-to-real data ratio in shipping policies. The higher the fraction of training data that is simulated yet still transfers, the faster the whole field moves, because simulation is the only exit that scales cheaply.
• Mean cycles between interventions (MCBI). Track how many task cycles a deployed robot completes before a human has to step in. This is the number that separates a demo from a business, and it is the one vendors are slowest to publish.
• Cross-embodiment transfer. The day one policy runs competently on several different robot bodies without per-robot retraining is the day the "robotics foundation model" thesis stops being a hope.
• First VLA policy through a safety audit. Watch for the first learned, stochastic policy certified for autonomous operation under ISO 10218 or ISO/TS 15066. That certification, not any demo, is the real unlock for human-shared work.
• Verified $/hour at a real site. Independent confirmation of a humanoid or mobile manipulator's cost per hour beating human labor on a sustained shift, not in a pilot press release.

What will not happen

• No reliable, affordable general-purpose home robot by 2030. The demos will be stunning; your house will not have one doing chores dependably this decade.
• Humanoids won't replace purpose-built automation where a simpler shape does the job better and cheaper, which is most of the time.
• Sim-to-real won't fully close. Domain randomization (Tobin et al., 2017) and simulators like NVIDIA Isaac Sim do more each year, but the reality gap is worst where robots earn their money: contact. You can only randomize what you can model; the long tail of friction, impact, and deformation you cannot keeps the real world's veto intact.
• Robotics won't have its "ChatGPT moment" the same way. Physical reality has no copy-paste and no infinite training data; progress stays lumpier and slower than software AI.

What it means for you

If you build or buy robots, the durable move is to invest in the fundamentals that survive the hype cycle, actuation, control, motion planning, and calibration, because the learned-policy layer on top keeps changing while the physics underneath does not. And get fluent with the AI tools now rewriting the development workflow; a roboticist who uses a model like Claude to scaffold code, generate sim scenarios, and reason through failure modes simply moves faster than one who doesn't.

The next decade of robotics rewards the people who can tell the demo from the deployment. Learn to see the gap.

Related flagships: the foundations behind all of this, the Robotics Canon, and how to actually skill up, Best Robotics Certifications & Courses. For the money side of the same story, read the robotics funding and capital cycle.

Key takeaways

• Software is the story; data is the constraint. VLA policies are replacing hand-coded pipelines, and the labs that win are the ones solving robot-action data collection across teleoperation, simulation, and human video at once.
• Structure decides the order. Robots ship first where the environment is structured and failure is recoverable. Warehouses and industrial cells scale; the unstructured home comes last, if at all this decade.
• Humanoids earn their keep only in narrow, structured work. Digit's warehouse totes are the honest benchmark. Cheap dev platforms like the Unitree G1 and certified production humanoids are different products; do not conflate them.
• Reliability is the moat, not intelligence. A 95% demo is a 3.5e-5 chance of a clean 200-cycle run. The jump to the parts-per-million world of industrial automation is a different discipline, and more data alone does not deliver it.
• Certification is the real long-term unlock. Fitting stochastic learned policies inside ISO 10218, ISO/TS 15066, ISO 13482, and IEC 61508 is the decade's hardest, least glamorous problem.
• Bet on the fundamentals. Actuation, control, motion planning, and calibration outlast every model generation because the physics underneath does not change.

FAQ

Q: Will humanoid robots be in homes within 10 years? Almost certainly not as reliable, affordable, general-purpose helpers. Expect impressive demos and real deployment in factories and warehouses, but the home is the hardest environment (unstructured, safety-critical, and unforgiving) and the economics and reliability won't be there for mass consumer adoption this decade.

Q: What's the biggest change coming in robotics? The software. Foundation models and vision-language-action policies are replacing hand-coded, task-specific pipelines, so robots increasingly learn general behavior instead of being programmed task by task. The binding constraint on this is data (there's no internet-scale dataset of robot actions), so data collection is the real frontier.

Q: Are humanoids the future of robotics? For a narrow set of tasks in human-built environments, yes, but the future of most automation is the right-shaped robot for the job, which is usually not humanoid. The human form is a powerful marketing and general-purpose argument and an engineering disadvantage for most specific tasks.

Q: What's the hardest unsolved problem in robotics? Reliable, data-efficient general manipulation in unstructured environments, and the data bottleneck behind it. Getting a robot to do one thing well is solved; getting it to do new things reliably without enormous task-specific data and engineering is the open problem the whole field is racing toward.