From Roomba to Atlas: The Smart Level Scale Explained, and Where Every Robot Falls

You own a robot. You probably do not think of it that way, but if there is a Roomba bumping around your living room, a smart speaker answering your questions, or a robot mower chewing through your backyard, you already live with machines that make decisions on their own. The question is: how smart are they, really? And where do they fall on the spectrum that ends with a humanoid robot walking into your office and asking what you need help with?

That is the problem the Smart Level scale was built to solve.

Every robot, smart home device, and autonomous vehicle tracked on this site gets a rating from 1 to 10. The scale measures something specific: the degree to which a machine can perceive its environment, make decisions, learn from experience, and interact with humans. It is not a quality score. A level 5 Roomba is an excellent vacuum. A level 6 Pepper is a mediocre conversationalist. The number tells you about capability, not about whether the thing is any good at its job.

This article walks through all ten levels, explains what separates each one from the next, and places real robots at every tier. By the time you reach the end, you will have a map of the entire robotics landscape, from the devices you already own to the machines that are coming for your workplace.

Why a scale matters

The robotics industry has a language problem. When Tesla says Optimus is “autonomous,” when Boston Dynamics shows Atlas doing backflips, and when iRobot says the Roomba j9+ uses “intelligent mapping,” they are all using the same vocabulary to describe wildly different capabilities. A vacuum that avoids chair legs and a humanoid that coordinates with humans on a factory floor both get called “smart,” but the distance between those two things is enormous.

The automotive industry solved a similar problem years ago. SAE International created the J3016 standard, which defines six levels of driving automation from Level 0 (no automation) to Level 5 (full automation). That scale gave consumers, regulators, and engineers a shared framework for talking about self-driving cars. Everyone knows what “Level 2” means, and it is very different from “Level 4.”

Robotics needed the same thing. Academic taxonomies exist, such as the autonomy frameworks published in Nature and the classification systems used by NIST, but they are written for researchers, not for someone trying to understand what their new robot can actually do. The Smart Level scale translates those concepts into a 1-to-10 rating that works for any machine, from a $30 smart bulb to a $2 million humanoid.

The full scale: all 10 levels

Before diving into each level in detail, here is the complete Smart Level scale in one view.

The gap between each level is not uniform. Moving from level 1 to level 2 is a small step. Moving from level 8 to level 9 is a giant leap. And moving from level 9 to level 10 might be the hardest problem in all of engineering. Let us walk through them one by one.

Level 1: Remote

Definition: Controlled entirely by a human operator with no autonomous behavior.

This is the simplest category. A level 1 robot does nothing on its own. Every action requires a human pressing a button, moving a joystick, or sending a command. There is no onboard intelligence. The machine is a tool, like a power drill that happens to be controlled wirelessly.

You will not find many consumer robots at this level today because even cheap toys include basic programming. But plenty of industrial and military robots still operate here. Bomb disposal robots that a technician drives via remote control, surgical robots where the surgeon controls every movement, and basic RC drones without autopilot all qualify as level 1.

The key distinction: if you stop giving commands, a level 1 robot stops doing anything.

Level 2: Reactive

Definition: Follows pre-programmed rules. Responds to simple triggers like bumping into walls.

Level 2 is where the robot starts doing something on its own, but only in the most basic way. It follows if-then rules hardcoded by its designers. If it hits a wall, it turns. If it detects an edge, it backs up. There is no map, no memory, no learning. Each moment is independent.

Example from the database: The Eilik desktop companion robot sits at level 2. It responds to touch, tilting, and proximity with pre-programmed animations and reactions. Poke it and it flinches. Tilt it and it flails. It is charming, but there is no understanding behind the behavior. Eilik does not learn your habits or remember yesterday. Every interaction starts from zero.

Early Roomba models (pre-iAdapt) also lived at this level. They bounced off walls in semi-random patterns, covering a room through sheer persistence rather than any spatial understanding.

Level 3: Scheduled

Definition: Operates on timers and routines. Can follow a set path or schedule without input.

At level 3, time enters the picture. The robot can be told “do this at 8 AM every Tuesday” and it will. It can follow a predetermined path or sequence. But it cannot react to surprises beyond basic obstacle avoidance, and it does not adjust its routines based on what it encounters.

Examples from the database: The LIFX A19 smart bulb (level 3) operates on schedules and scenes. Set it to dim at 10 PM and brighten at 7 AM, and it will follow that routine faithfully. The Aiper Seagull Pro pool cleaner (level 3) follows programmed cleaning routes. The Dolphin Nautilus CC Plus (level 3) does the same. The CATLINK Scooper Luxury Pro litter box (level 3) runs cleaning cycles on fixed intervals.

These devices are reliable and useful, but they do not understand context. Your smart bulb does not know you came home early. Your pool cleaner does not know the leaves are heavier this week. They follow the script.

Level 4: Aware

Definition: Maps and understands its environment. Avoids obstacles and adapts routes in real time.

This is the level where robots start to “see.” A level 4 machine uses sensors, cameras, or lidar to build a model of its surroundings and navigate accordingly. It does not just bounce off obstacles. It detects them in advance and plans around them. It builds a map and knows where it is on that map.

Examples from the database: Robot lawn mowers like the Husqvarna Automower 450X, Mammotion LUBA 2 AWD, and EcoFlow Blade all sit at level 4. They use GPS, cameras, or boundary sensors to map your yard and mow systematically rather than randomly. Smart locks like the August Wi-Fi Smart Lock and Yale Assure Lock 2 are level 4: they know who is at the door (geofencing, auto-unlock) and adapt their behavior based on proximity. The Litter-Robot 4 uses weight sensors and cat detection to time its cleaning cycles around actual usage rather than fixed schedules. The Philips Hue starter kit (level 4) detects ambient light and adjusts brightness accordingly.

The dividing line between level 3 and level 4 is environmental awareness. A level 3 device follows instructions. A level 4 device perceives its surroundings and adjusts its behavior in real time.

Level 5: Adaptive

Definition: Learns from experience over time. Adjusts behavior based on patterns and preferences.

Level 5 is where things get interesting, because this is where learning begins. A level 5 robot does not just perceive its environment. It remembers what happened before and changes its behavior based on that memory. It gets better at its job over time. It recognizes patterns in your life and adapts to them.

The flagship example: The iRobot Roomba j9+. This is the robot most people think of as “just a vacuum,” but it is doing more than you might realize. The j9+ builds a persistent map of your home using vSLAM (visual simultaneous localization and mapping). It remembers which rooms it has cleaned and which it has not. It learns your schedule and suggests optimal cleaning times. Its PrecisionVision Navigation uses a front-facing camera to identify and avoid specific objects, cables, pet waste, shoes, socks, and it gets better at this over time through firmware updates and cloud-based machine learning.

That last part is what makes it level 5 rather than level 4. A level 4 robot avoids obstacles. A level 5 robot learns to recognize specific types of obstacles and adjusts its strategy based on accumulated experience.

Other level 5 machines include the Roborock S8 MaxV Ultra (AI-powered object recognition that improves with use), the Ecovacs Deebot X2 Omni, the Samsung Bespoke Jet Bot Combo AI, the Amazon Echo (Alexa learns your routines), the Google Nest Hub (adapts to your habits), and the Apple HomePod. The iRobot Braava Jet m6 mopping robot also sits at level 5, coordinating with the Roomba to mop after vacuuming based on learned room layouts.

Vector 2.0, the small companion robot from Digital Dream Labs, also reaches level 5. It recognizes faces, responds to voice commands, and develops different behavioral patterns over time based on interaction history. It is the simplest robot in the database that starts to feel like it “knows” you.

Level 6: Interactive

Definition: Recognizes faces, understands speech, and holds basic conversations. Socially aware.

Level 6 is where robots start to feel like characters. The defining feature is social intelligence: the ability to recognize individual humans, understand natural language, carry on a conversation, and respond to emotional cues. A level 6 robot is not just processing your words for commands. It is trying to understand your intent and respond in a way that feels natural.

Examples from the database: SoftBank’s Pepper (level 6) was one of the first commercial robots designed primarily for social interaction. It detects emotions through facial analysis, holds scripted but convincing conversations, and adapts its tone to the situation. SoftBank’s NAO (also level 6) is widely used in autism therapy and education because of its ability to engage children in social interaction. Hanson Robotics’ Sophia (level 6) can hold extended conversations, recognize returning visitors, and generate contextually appropriate facial expressions. Engineered Arts’ Ameca (level 6) takes expressive interaction even further with hyper-realistic facial movements.

Sony’s Aibo (level 6) is fascinating because it achieves social intelligence in a non-humanoid form. The robotic dog recognizes family members, responds differently to each person, develops a distinct “personality” over time, and uses body language (ear positions, tail wagging, posture) to communicate emotions. It does not talk, but it is socially aware in a way that level 5 devices are not.

Amazon Astro (level 6), the home robot, patrols your house autonomously, recognizes family members by face, and carries on basic conversations using Alexa. The Starship and Serve delivery robots (both level 6) navigate sidewalks autonomously while communicating with pedestrians and customers.

The Ai-Da art robot, which creates original artworks based on visual input and can discuss its creative process, also sits at level 6. It is one of the stranger entries in the database but demonstrates an important point: social intelligence is not limited to utilitarian tasks.

Level 7: Autonomous

Definition: Navigates complex environments independently. Makes real-time decisions without human input.

Level 7 is where the word “autonomous” actually starts to mean something. At this level, a robot can be dropped into a complex, unstructured, changing environment and figure out what to do without any human help. It does not need a map provided in advance. It does not need someone monitoring it. It handles the unexpected.

The defining example: Boston Dynamics Spot (level 7). The quadruped robot navigates construction sites, power plants, oil rigs, and disaster zones. It walks up stairs, picks its way across rubble, opens doors, and recovers from slips and falls, all without a human driving it. Spot does not just follow a path. It evaluates terrain in real time, selects footholds, and adapts its gait to conditions that its designers could never have predicted in advance.

Honda’s ASIMO (level 7), though now retired, was one of the first humanoid robots to achieve this tier. It could walk on uneven surfaces, climb stairs, avoid moving obstacles, and navigate through crowds, all autonomously. ASIMO could also run, kick a soccer ball, and pour a drink, demonstrations of physical autonomy that seemed almost magical when they debuted in the early 2000s.

Xiaomi’s CyberOne (level 7) uses real-time environment perception and can navigate indoor spaces while avoiding dynamic obstacles. Engine AI’s SE01 (level 7) and Samsung’s Bot Handy concept (level 7) also sit at this tier, demonstrating autonomous navigation in home and retail environments.

In the autonomous vehicle space, Gatik’s box trucks (level 7) and May Mobility’s shuttles (level 7) operate on fixed routes but handle complex traffic situations autonomously, making them level 7 rather than level 4.

The difference between level 6 and level 7 is focus. A level 6 robot is smart about people. A level 7 robot is smart about the physical world. Spot cannot hold a conversation, but it can traverse an environment that would defeat any level 6 machine.

Level 8: Skilled

Definition: Performs multi-step physical tasks. Manipulates objects, uses tools, and learns new skills from demonstration.

Level 8 is the tier where robotics gets serious about replacing human labor. At this level, robots do not just navigate. They work. They pick up objects, use tools, assemble components, sort packages, and learn new tasks by watching humans do them first. The hands become as important as the legs.

This is the most crowded tier in our database, and for good reason. It is where the humanoid robot industry is actually competing right now.

The affordable champion: Unitree’s G1 (level 8) starts at roughly $16,000, making it the cheapest full-size humanoid robot you can buy. It performs multi-step manipulation tasks, uses tools when equipped with the right end-effectors, and learns new movements through demonstration. Its 23 to 43 degrees of freedom give it enough dexterity for warehouse sorting, light assembly, and research applications.

The corporate bet: Tesla’s Optimus Gen 2 (level 8) is being trained in Tesla’s own factories, learning to sort battery cells, move boxes, and navigate warehouse aisles. Tesla’s advantage is data: every Optimus unit feeds training data back to a central model, which means every new unit benefits from the experience of all previous units.

The warehouse worker: Agility Robotics’ Digit (level 8) is already deployed at Amazon fulfillment centers, picking up totes and moving them between locations. Digit demonstrates one of the key level 8 capabilities: the ability to handle diverse objects without being specifically programmed for each one.

Other level 8 machines include Sanctuary AI’s Phoenix (which uses dexterous hands for fine manipulation), Apptronik’s Apollo (designed for automotive and logistics), Fourier’s GR-2 (rehabilitation and general purpose), NASA’s Robonaut 2 (operating on the International Space Station), UBTECH’s Walker S2, AgiBot’s A2, Kepler’s Forerunner K2, XPENG Iron, Robot Era STAR1, LimX Dynamics Oli, and NEURA’s 4NE-1.

In the autonomous vehicle space, Tesla’s Cybercab, Pony.ai’s PonyPilot+, WeRide’s S5, Nuro’s R3 delivery vehicle, Aurora’s autonomous trucks, and TuSimple’s autonomous semis all qualify as level 8. They perform complex multi-step driving tasks in mixed traffic but operate within defined domains rather than handling any situation a human driver could.

The level 8 gap: what separates the best from the rest

Not all level 8 robots are equal. There is a meaningful difference between a robot that can pick up a box in a controlled demo and one that can do it reliably for eight hours in a warehouse. The industry informally distinguishes between “narrow 8” (performing specific tasks very well) and “broad 8” (performing many different tasks adequately).

Digit and Optimus are currently “narrow 8” machines. They excel at specific tasks they have been trained for. The G1, with its open SDK, is being pushed toward “broad 8” by its research community, which uses it for everything from cooking experiments to parkour.

The frontier of level 8 is learning speed. How many demonstrations does a robot need before it can perform a new task? Current machines typically need dozens to hundreds of examples. Getting that number down to single digits is what will push the best level 8 robots toward level 9.

Level 9: Collaborative

Definition: Works alongside humans as a partner. Understands context, intent, and can coordinate on shared goals.

Level 9 is where robots stop being tools and start being partners. The defining capability is not any single skill but the integration of many skills into coherent, context-aware behavior. A level 9 robot understands what you are trying to accomplish, anticipates what you need next, and coordinates its actions with yours in real time. It does not wait for instructions. It contributes.

The pinnacle of humanoid robotics today: Boston Dynamics Atlas (level 9) represents the most advanced integration of physical capability and autonomous decision-making in any humanoid robot. The electric version of Atlas, unveiled in 2024, can manipulate heavy objects, navigate unpredictable terrain, recover from disturbances, and work alongside human operators in industrial settings, all while coordinating its actions with the humans around it.

What pushes Atlas from level 8 to level 9 is collaboration. In demonstrations at Hyundai’s manufacturing facilities, Atlas has shown the ability to understand the intent of human co-workers, hand off objects at the right moment, adjust its pace to match a human partner, and modify its approach when a shared task does not go as planned. This is not just reactive behavior. It requires a model of what the other agent (the human) is thinking and planning.

Figure 03 (level 9) also reaches this tier. Figure AI’s third-generation humanoid combines their Helix AI platform with improved hardware to achieve fluid human-robot collaboration. The key advancement is conversational task coordination: a human can explain a multi-step task in natural language, and Figure 03 will execute it while asking clarifying questions and flagging problems. It is the closest any humanoid has come to working like a new employee on their first day.

In the autonomous vehicle domain, several systems have reached level 9. Waymo One (level 9) operates a fully autonomous taxi service with no safety driver across multiple cities, handling the full complexity of urban driving including construction zones, emergency vehicles, and unpredictable pedestrians. Baidu’s Apollo RT6 (level 9) does the same in Chinese cities. Cruise’s Origin (level 9) and Zoox’s robotaxi (level 9) represent purpose-built autonomous vehicles designed from the ground up for driverless operation.

These level 9 autonomous vehicles are instructive because they demonstrate what “collaborative” means in practice. A Waymo does not just drive. It communicates with passengers through screens and speakers, coordinates with other vehicles, yields to emergency services, and adapts to the unwritten social rules of driving that vary from city to city. It is participating in a shared human system, not just executing a driving algorithm.

Level 10: General

Definition: Human-equivalent reasoning and adaptability across any task or environment.

No robot on this site, or anywhere on Earth, is rated level 10. Not one.

Level 10 is the theoretical ceiling of the scale: a machine with human-equivalent reasoning, creativity, common sense, and adaptability. A level 10 robot could walk into any environment, assess any situation, learn any task, hold any conversation, and make any judgment call that a competent human adult could make. It would not be specialized. It would be general.

This is what the AI research community calls Artificial General Intelligence (AGI) embodied in a physical form. It is the explicit long-term goal of companies like Figure AI, Tesla, and 1X Technologies. It is also the implicit goal of the entire humanoid robotics industry, because general-purpose robots are the only ones that justify the enormous investment being poured into this field.

The honest assessment: level 10 is likely decades away, if it is achievable at all. The problems that remain, common-sense reasoning, transfer learning across radically different domains, genuine creativity, understanding of abstract concepts, are not engineering problems that can be solved with better hardware. They are open research questions that the AI field has been working on for 70 years.

Goldman Sachs projects the humanoid robot market will reach $38 billion by 2035. That projection assumes level 8 and level 9 robots becoming commercially viable at scale. It does not require level 10. The near-term future of robotics is not about building a mechanical human. It is about building machines that are excellent at specific, valuable tasks and that keep getting better.

The complete landscape: every level, every robot

Here is a summary view of where every robot category falls on the Smart Level scale, based on the machines currently tracked on this site.

What the scale reveals about the industry

When you step back and look at the full scale, several patterns emerge.

The consumer ceiling is level 5. Most people interact with robots at level 5 or below. Your vacuum, your speaker, your security camera. These devices are genuinely intelligent in narrow ways, and they are getting better every year. But they are not going to surprise you. They are not going to do something you did not expect.

The commercial floor is level 8. For a humanoid robot to be commercially useful in a factory, warehouse, or hospital, it needs to reach level 8 at minimum. That is why every humanoid company is racing to this tier. Below level 8, a humanoid is a research project. At level 8, it is a product.

The social robots are stuck. Level 6 machines like Pepper, Sophia, and NAO have not meaningfully advanced in years. Social intelligence turned out to be harder to monetize than anyone expected. The path from “entertaining demo” to “valuable product” is unclear for machines whose primary skill is conversation. The recent advances in large language models could change this, but so far, the level 6 robots have not been the ones benefiting from the AI boom.

Autonomous vehicles are ahead of humanoids. The smartest autonomous vehicles (Waymo, Zoox) are rated level 9, the same as the smartest humanoids. But they have been at level 9 for longer and are further into commercial deployment. This makes sense: driving is a more constrained problem than general manipulation. But it also means that the robotics industry’s most advanced real-world AI is not in a humanoid at all. It is in a minivan.

How to use the Smart Level when shopping

If you are a consumer trying to decide what to buy, the Smart Level gives you a quick way to set expectations.

Levels 3-4 are “set it and forget it” devices. Lawn mowers, smart locks, basic lighting. They do one thing reliably with minimal attention from you.

Level 5 devices are “smart but single-purpose.” They learn your habits and improve over time, but they are locked to one domain. Your Roomba will never answer a question. Your Echo will never clean the floor.

Level 6 devices are “social companions.” If you want something that feels interactive and personality-driven, this is the tier. But set your expectations for utility accordingly. Pepper is fun at a trade show. It is not going to do your dishes.

Levels 7-8 are “the professional tier.” These machines are expensive, specialized, and powerful. Unless you are a researcher, developer, or enterprise buyer, you are probably not shopping here yet. But check back in a few years.

Level 9 is “the frontier.” You can ride in a Waymo today, which means you can already experience level 9 intelligence as a passenger. For humanoids, level 9 remains an industrial and research technology.

The road ahead

The Smart Level scale is not static. As the technology advances, the bar for each level rises. What qualified as level 5 in 2020 might only qualify as level 4 today, because the industry’s collective understanding of what “adaptive” means has matured. We periodically review and recalibrate the ratings for every robot in the database.

The most exciting near-term development is the rapid advancement of level 8 humanoids. A year ago, most of these machines could barely walk. Today, they are sorting packages, assembling components, and learning new tasks in hours instead of weeks. The improvement curve is steep, and it is being driven by the same foundation model revolution that gave us ChatGPT and its successors.

Level 10 remains aspirational. But the ground between level 8 and level 9 is being covered faster than most people expected. The robots are getting smarter. The question is not whether they will reach the upper levels of the scale but how quickly, and what we will do when they get there.

The Roomba you own today is a level 5. The humanoid that might work in your office in 2030 will be a level 8 or 9. The distance between those two numbers is the distance between a machine that cleans your floor and a machine that understands what you need before you ask for it. That is the story the Smart Level scale tells, and it is a story that is still being written.