Signal Atlas

Research library

Robots & autonomous systems

Machines that sense, decide and act, from factory cobots to humanoids, drones and human–machine teams. What 6G adds: reliable control links and shared sensing for machines that move among people.

Of 44 findings in this theme, 11 are real today, 14 designed and 19 forecast.

Real today 11

  • A military 'kill box' is an area for coordinating fires. In US joint doctrine a kill box is a three-dimensional area used to coordinate air and surface weapon fire, with 'Blue' and 'Purple' types. It is a battlefield coordination tool from the late 1980s, not a list of houses.
  • About 7,000 full-size humanoids sold worldwide in 2025. The International Federation of Robotics counted roughly 7,000 full-size humanoids (over 140 cm) sold in 2025. In the same year almost 250,000 professional service robots and about 34 million consumer robots, mostly vacuums and mowers, were sold.
  • Ageing populations make care automation more attractive. Fewer workers and more older people raise care demand and the pull to automate. Automation can either help older or scarce workers do more, or simply cut headcount; design decides which.
  • BINA48 animated one person's archive in a robot. A 2012 TEDxHarlem talk featured BINA48, an android head modelled on Bina Aspen Rothblatt, answering questions from a database built from interviews and other material about her (some sources say it blends several people's information). The presenters admitted it was imperfect.
  • Cheap autonomous drones are spreading fast. The U.S. Replicator effort aims to field large numbers of expendable autonomous systems, the war in Ukraine is driving rapid AI-assisted drone development, and the U.S. updated its autonomy-in-weapons directive in 2023.
  • Drones already inspect telecom towers on their own. Telecom operators use drones to inspect radio towers, and some systems photograph, analyse and report without a pilot steering them. Many countries still require an on-site human pilot, which limits savings.
  • Four military programmes she names are real but separate. ATLAS (aided tank targeting), Skyborg (AI drone wingman), NTS-3 (navigation satellite) and UK Skynet 6 (military satellites) are separate real programmes. The sources do not show them forming a single system or involving body networks.
  • General robots are still clumsy; control and cost block. Forethought describes current general-purpose robots as very clumsy outside controlled settings, even though human-guided surgical robots can be extremely precise. It names control software and cheap manufacturing as the two main barriers.
  • Hospital delivery robots already work on ordinary Wi-Fi. Diligent Robotics says its Moxi hospital robot needs no new infrastructure and connects over the hospital's existing Wi-Fi. 1X lists Wi-Fi, Bluetooth and 5G for its NEO home robot.
  • Living robots made from frog cells. Wyss Institute researchers built 'xenobots', tiny living machines assembled from frog cells that can move, self-repair and, in one study, reproduce.
  • Project Maven helps find targets; analysts decide. Project Maven is a Pentagon AI programme that began in 2017 to analyse drone and satellite imagery. Google left the work in 2018, and the programme now supports targeting while human analysts review its results.

Designed, not deployed 14

  • 3GPP's 6G study lists robot fleets and sensing. The global standards body 3GPP finished its Release 20 study of 6G use cases (TR 22.870) in early 2026. Robot examples include the network sensing surroundings to guide factory robots and robot fleets sharing data in real time.
  • 6G could help robots plan around radio coverage. Studies feeding 3GPP describe humanoid and other robots using 6G for search-and-rescue mapping, routes that avoid coverage gaps, sensing an object's material before touching it, and coordinated warehouse fleets.
  • 6G targets connected driving at up to 1,000 km/h. ITU names network-assisted automated driving and devices collaborating autonomously as typical AI-and-communication uses, and sets a research target for staying connected at 500-1,000 km/h.
  • Autonomy that reads its human teammate. U.S. Army Research Laboratory scientists have proposed sensing an individual soldier's brain and body signals so machines can adapt, for example a 'shared control' vehicle that alerts the driver or hands back control depending on their state. Drowsiness detection was a planned simulator experiment, not a result.
  • Figure 03 uploads its data wirelessly at 10 Gbps. Figure's 2025 home-focused humanoid charges through its feet and uses a 10 Gbps millimetre-wave link to upload large amounts of data so the whole fleet can keep learning.
  • Four levels of machine autonomy. IEEE's Symbiotic Autonomous Systems white papers separate remote control, teleoperation, semi-autonomous operation and full autonomy, following earlier NIST work on autonomy levels. Real systems move between these modes.
  • Four ways humans stay in charge of machines. A system can wait for human approval, act while a human supervises, act alone under human-set policy, or treat the person's body signals as one more input. Real override also needs time to react, a control path separate from the failing software, and a safe state afterwards.
  • In-store 5G planned to run fleets of shop robots. AT&T and Badger Technologies announced using in-store 5G with edge computing so many autonomous robots could roam a shop for inventory and shelf restocking. A private 5G network was also pitched as harder to hack than Wi-Fi.
  • Machines infer intent and context, not feelings. Autonomous systems track surroundings, battery, task progress and guessed human intent, and may treat gaze, hesitation or body signals as commands. The white paper calls this awareness only in a functional sense.
  • Network sensing could let people and robots share floors. Nokia Bell Labs describes a network that bounces radio signals off objects to see what is there and how it moves, so factories could let people and industrial robots work side by side safely.
  • Robot swarms, vehicle fleets and smart cities coordinate. Autonomous systems can cooperate through a central controller, distributed rules, or emergent group behaviour. Examples include robot swarms, vehicle fleets that share learned models, IoT environments and smart cities.
  • Robots near people need sensing skins and active safety. As robots move into homes and workplaces, they must watch their surroundings and change how they move when a person touches or bumps them. Flexible sensor skins on robot arms are one proposed way to do this.
  • Robots spread slower than software because hardware costs. Unlike AI software, a robot must work safely in messy physical spaces and pay back equipment, installation, safety approval and maintenance. So physical automation usually arrives piece by piece, changing tasks before whole jobs disappear.
  • Symbiotic autonomous systems: interact, evolve, decide. IEEE's 2017 white paper defines symbiotic autonomous systems by three abilities: rich interaction with people and machines, self-evolution through learning and updates, and autonomous decision-making. Humans and machines continually adapt to each other.

Forecast 19

  • 2019 forecast: assistive and humanoid robot markets to multiply. An IEEE white paper from 2019 put the assistive robot market at $359 million in 2018, expected to reach $1.2 billion by 2024. It said humanoids, popular in East Asia, could exceed $5 billion.
  • A $20,000 home humanoid is on pre-order. 1X opened pre-orders for its NEO home robot in November 2025: $20,000 to buy or $499 a month, with a $200 refundable deposit. It promises US deliveries in 2026 and other markets in 2027.
  • A billion humanoids a year needs huge cost drops. A 2025 Forethought paper estimates roughly 10,000 humanoids are built per year today. If costs fell with scale the way solar panels did, reaching a billion a year could cut unit cost by about 97 percent.
  • A working humanoid could pay for itself in months. The same paper argues a humanoid costing about $100,000 that replaces a $50,000 manual job, and works close to around the clock, would earn back its price in under six months, so robot numbers could double very fast.
  • AI-built robots could multiply fast after an intelligence explosion. A 2025 Forethought paper argues that if AI speeds up research, robots building robots could drive an 'industrial explosion', with robot numbers doubling in months. It estimates about 10,000 humanoids are made today and sketches a path to a billion robots a year.
  • Android built from one person's life data. Bina48, shown at TEDxHarlem in 2012, is a talking robot head built from interviews and records of a real woman, part of a project testing whether rich personal data archives could one day recreate a personality in avatars or robots.
  • Canada may see office job pressure alongside care shortages. Because AI hits digital work quickly while robots move through capital and safety limits, Canada could face pressure in office and professional jobs while still short of workers in care and construction.
  • Cars likely become AI co-pilots before full autopilots. Isaac Arthur expects safety features and AI co-pilots that take over when a driver is distracted to win first, with strong social or legal pressure to let computers drive arriving in roughly 20 to 30 years. He expects automated freight to come slower than some predict.
  • Cobots and digital twins spread through Industry 4.0. Factories lead the adoption of autonomous systems: robots and collaborative robots work beside people, and digital twins of equipment were becoming the norm. One forecast put the cobot market at about $12.3 billion by 2025, with exoskeletons growing too.
  • Cobots only pay off when work and training change. Using collaborative robots well means reorganising the workflow and retraining staff to work alongside them. The white paper calls this one of industry's biggest challenges, while quoting a $12.3 billion cobot market forecast for 2025.
  • Commentators place a billion robots in the 2040s. Commentator David Shapiro expects robots to take off in the mid-to-late 2030s, threatening even skilled trades, and a billion robots closer to 2040-2050. An interview guest likewise floated a billion humanoids within the next decades.
  • Figure 03 is designed for homes and mass production. Figure's 2025 robot has washable soft covers, foam at pinch points and self-docking for home use. Its factory starts at up to 12,000 robots a year, aiming for 100,000 over four years.
  • Fully self-driving private cars sit beyond the 2020s. The IEEE paper did not expect full (level 5) autonomy for private cars in its timeframe, though it cited forecasts of about 700,000 level-5 vehicles sold in 2030 and 100,000 robotaxis by 2022. Self-driving cars were expected to become electric and shift mobility from ownership to service later on.
  • Humanoids first, because the world fits human bodies. Because tools, buildings and appliances are made for human bodies, humanoids may be in demand first. Over time special-purpose machines and redesigned spaces could be more efficient than human-shaped robots.
  • Mind files might move between avatars and robots. The LifeNaut project behind BINA48 hypothesised that a rich enough life archive could drive a personality model that moves between computers, avatars, robots and, far later, new kinds of bodies.
  • Podcast voices claim humanoids do most physical work soon. Several talks the owner saved claim humanoids will match human dexterity within about five years and could take over most physical labour around 2030. They point to many companies, including Tesla, racing to build them.
  • Social robots sell more but lack a killer use. The same white paper reported social and entertainment robot sales doubling from 1.7 million in 2015, heading toward 7.4 million by 2025. It warned that without a clear killer use they would stay marginal.
  • Thousands of AI and robot helpers per person, maybe. In a fast AI-driven growth scenario, the paper imagines each person could have access to thousands of AI and robot assistants, if people want that.
  • Trades: low AI exposure, higher robot exposure in Canada. Statistics Canada estimated in January 2026 that about 20 percent of employees in certified journeyperson trades are at high risk of automation-related job transformation, versus 13 percent in other occupations, even though most trades are less exposed to AI.

Other themes