Humanoid Robot Features Explained: What to Look For When Buying

Spec sheets for humanoid robots can read like an engineering thesis. Degrees of freedom, LiDAR resolution, battery watt-hours — manufacturers love throwing numbers at you. But which of those numbers actually matter when you are buying a robot for your home, and which are pure marketing fluff?

We have spent months testing, comparing, and living with consumer humanoid robots. In this guide, we break down every major feature category so you can separate the genuinely useful specs from the ones that only impress on paper. If you are new to the market entirely, bookmark this page — you will be coming back to it.

Degrees of freedom (DOF): what the numbers mean

Degrees of freedom refers to the number of independent axes a robot’s joints can move along. A single hinge joint, like your elbow, provides one DOF. A ball-and-socket joint, like your shoulder, provides three. Add all those axes together across the full body and you get the total DOF count.

Current consumer humanoids typically advertise between 33 and 44 DOF. That sounds like a wide range, but here is the nuance: more DOF does not automatically mean a more capable robot for home use. A robot with 44 DOF that dedicates twelve of those to hyper-articulated fingers may look spectacular folding origami in a demo video, but if the control software cannot coordinate those joints reliably, you end up with a machine that fumbles a coffee mug.

What actually matters is how many useful, well-controlled degrees of freedom the robot has. For household tasks — picking up objects, opening doors, loading a dishwasher — you want at least 6 DOF per arm (shoulder, elbow, and wrist articulation) and a hand with enough dexterity to manage a reliable pinch grip and a power grip. Anything beyond that is a bonus, but only if the software can keep up. We discuss specific models and their real-world dexterity in our head-to-head comparison.

Battery life: capacity versus real-world runtime

Manufacturers quote battery capacity in watt-hours (Wh). You will see figures ranging from around 700 Wh to over 1,500 Wh across current models. But raw capacity tells you very little without context.

A robot standing still, maintaining balance, and running its AI stack might draw 150–200 W. Walking on a flat surface bumps that to 300–400 W. Active manipulation — carrying objects, reaching, gripping — adds another 50–100 W on top. So a 1,000 Wh battery does not give you five hours of useful runtime; it gives you roughly two to three hours of mixed household activity before the robot needs to find its charging dock.

What to look for: prioritise real-world runtime estimates over raw Wh numbers. Ask whether the manufacturer’s quoted runtime assumes idle standing or active use. Check whether the robot can autonomously return to its charger (most 2026 models can). And if the robot offers hot-swappable battery packs, that is a genuine advantage for longer tasks. Our setup guide covers charging placement and power management in detail.

AI capabilities: on-device versus cloud processing

This is where the consumer robotics market splits sharply. Some robots handle most of their intelligence on-device using dedicated neural processing units (NPUs), while others rely heavily on cloud-based large language models for anything beyond basic navigation.

On-device AI means lower latency, continued operation if your broadband drops, and better privacy. The trade-off is that on-device hardware has limited compute, so the models tend to be smaller and less capable for open-ended conversation or complex reasoning tasks.

Cloud-based AI unlocks more powerful LLM integration — think natural language task planning, contextual understanding of ambiguous requests, and the ability to improve over time through model updates. The downsides are latency (a noticeable pause before complex responses), dependence on a stable internet connection, and the reality that your voice commands and camera feeds are being processed on remote servers.

The best current approach, and the one we recommend looking for, is a hybrid architecture. The robot handles safety-critical functions (balance, obstacle avoidance, emergency stops) entirely on-device, while offloading higher-level planning and conversation to the cloud. Ask the manufacturer what happens when the internet goes down — if the robot becomes a glorified statue, that is a red flag.

Sensors: the robot’s eyes, ears, and spatial awareness

A humanoid robot’s sensor suite determines how well it perceives and navigates your home. Here are the key types:

LiDAR creates precise 3D maps of the environment. Essential for reliable navigation around furniture, pets, and children. Look for 360-degree coverage rather than forward-facing only.
Depth cameras (structured light or time-of-flight) provide close-range 3D perception critical for manipulation tasks. Most robots mount these in the head and wrists.
RGB cameras handle object recognition, face identification, and general scene understanding. Resolution matters less than the quality of the vision AI processing the feed.
IMU (inertial measurement unit) tracks the robot’s own orientation and acceleration. This is fundamental to balance. All humanoids have IMUs, but higher-quality units with faster update rates produce smoother, more stable movement.
Touch and force sensors in the hands and fingers allow the robot to gauge how hard it is gripping. Without these, a robot that can pick up an egg will eventually crush one.
Microphone arrays enable voice interaction and sound localisation. Look for at least four microphones in a far-field array configuration if you want reliable voice commands from across a room.

For home use, we would rank sensor priorities as: LiDAR and depth cameras first (navigation safety), then force/touch sensors (manipulation reliability), then everything else.

Manipulation: hands, grips, and real-world dexterity

Robot hands broadly fall into three categories in the consumer market:

Parallel grippers — two or three fingers that open and close. Simple, robust, and reliable for picking up most household objects. Limited dexterity for tasks requiring finger independence.
Multi-fingered hands (4–5 fingers) — closer to human anatomy. Greater potential dexterity, but significantly more complex to control. Software quality is the bottleneck here.
Soft or compliant grippers — flexible materials that conform to object shapes. Excellent for delicate items but less precise for tool use.

Grip strength is quoted in newtons or kilogrammes-force. For household tasks, 20–40 N of pinch grip is adequate for most objects. More important than peak grip strength is controllability — can the robot reliably hold a glass of water without either dropping it or shattering it?

Dexterity benchmarks are still maturing in the industry, so we recommend watching independent demo videos of the specific tasks you care about rather than trusting spec-sheet claims.

Locomotion: walking, terrain, and fall recovery

Walking speed for consumer humanoids currently ranges from about 1.2 to 2.5 metres per second — roughly a slow to moderate human walking pace. For home use, speed is far less important than stability and terrain handling.

Key locomotion questions to ask:

Can it handle common household transitions? Carpet to hardwood, over door thresholds, across rugs.
Can it manage stairs? Some models can descend stairs; fewer can ascend them reliably. If you have a multi-storey home, this matters enormously.
What is the fall recovery capability? Every robot will eventually stumble. The critical question is whether it can get back up autonomously and whether it fails safely (controlled descent rather than a rigid topple). A robot that crashes like a felled tree is a liability around children and pets.

Look for robots that have been demonstrated on varied surfaces, not just polished exhibition floors. Real homes have clutter, uneven flooring, and unexpected obstacles.

Connectivity and smart home integration

At minimum, you want dual-band Wi-Fi 6 and Bluetooth 5.0. But connectivity goes beyond raw wireless specs:

Smart home ecosystem support — does the robot integrate with Matter, Home Assistant, Google Home, or Apple HomeKit? The more ecosystems supported, the more useful the robot becomes as a central home automation hub.
Local network API — can you send commands to the robot over your local network without routing through the manufacturer’s cloud? This matters for both privacy and reliability.
Zigbee/Thread radio — some robots include these for direct communication with smart home devices. A nice-to-have, not essential if your existing smart home hub already covers this.

If you are building a connected home setup, check our best humanoid robots roundup for integration compatibility details.

Payload capacity

How much weight can the robot carry? Quoted payload capacities for consumer models typically range from 3 kg to 15 kg per arm, but these numbers usually refer to maximum static load with the arm fully extended — a best-case scenario.

For practical home use, think about what you actually need carried: a basket of laundry (3–5 kg), a bag of shopping (5–8 kg), a toddler (definitely not — no current consumer robot is rated for that, nor should you trust one to be). A robot with 5 kg of reliable, dynamic payload per arm will handle the vast majority of household carrying tasks without issue.

Noise levels

This is a spec that almost nobody talks about, but we think it matters enormously for a machine that shares your living space. Servo motors, cooling fans, and joint actuators all produce noise. Some robots hum along at a tolerable 40–45 dB (roughly the level of a quiet library), while others produce a persistent 55–60 dB whine that becomes genuinely irritating.

Ask for noise measurements during active movement, not idle standing. If the manufacturer does not publish dB levels, treat that as a warning sign.

Software ecosystem: SDK, apps, and updates

Hardware gets you through the door; software determines whether the robot is still useful in twelve months. Here is what to evaluate:

Companion app quality — is it polished and regularly updated, or does it feel like an afterthought? Check app store reviews, particularly recent ones.
SDK and developer access — if you want to customise behaviours or integrate with your own systems, an open SDK is invaluable. Some manufacturers lock everything behind proprietary tools.
Update frequency and commitment — how often does the manufacturer push software updates? Are those updates adding new capabilities or just patching bugs? A robot that received its last firmware update six months ago is a robot the manufacturer has moved on from.
Community size — a healthy user and developer community means more third-party skills, faster bug reports, and better collective knowledge.

The software ecosystem is, frankly, where most of the value lies. A mechanically average robot with brilliant software will outperform a beautifully engineered robot running mediocre code.

What actually matters for home use

Let us cut through the noise. If you are buying a humanoid robot for your home in 2026, here is our prioritised list of what to care about:

Software quality and update commitment — this determines the long-term trajectory of your purchase more than any hardware spec.
Sensor suite — reliable navigation and manipulation depend on good perception.
Real-world battery runtime — not quoted capacity, actual hours of mixed use.
Manipulation reliability — can it consistently handle everyday objects without drama?
Noise levels — you are living with this thing.
Smart home integration — the robot should enhance your existing setup, not exist in isolation.
Fall safety and recovery — especially important in homes with children or elderly family members.

Things that matter far less than manufacturers would have you believe: peak walking speed, total DOF count (beyond a practical minimum), maximum payload (you are not running a warehouse), and raw computing benchmarks.

For model-specific recommendations based on these priorities, see our best humanoid robots for UK homes guide. And if budget is a primary concern, our 2026 UK price guide breaks down what you can expect to spend across different tiers.

The humanoid robot market is moving fast, and spec sheets will keep getting more impressive. But the buyers who end up happiest are the ones who focus on real-world reliability over headline numbers. Buy the robot that works well today and has a manufacturer committed to making it work better tomorrow.