Service robotics is where robotics leaves the environment built for it.
What changes when the environment is not yours
An industrial cell is engineered: fixed lighting, known parts, controlled access, trained people, everything in a defined position.
A hospital corridor, a supermarket aisle, a hotel lobby or a warehouse during peak has none of that.
People are unpredictable and are not trained to behave around the robot. Children, distracted adults, people with limited mobility, people who deliberately obstruct it.
The layout changes. Furniture moves, boxes appear, doors close, floors get cleaned and become reflective.
Conditions vary. Lighting, noise, surfaces, spills, weather near entrances.
Objects are arbitrary. Not a known part in a known orientation, but whatever happens to be there.
Failure is public. A robot stuck in a factory is an availability statistic. A robot blocking a hospital corridor is an incident with witnesses.
Everything a factory removes to make robotics tractable is exactly what a service environment puts back.
What is deployed and what is a demonstration
Worth separating clearly, because this is the direction where the gap is widest.
Commercially deployed at meaningful scale. Warehouse mobile robots moving goods or shelves. Floor cleaning robots in retail and commercial buildings. Hospital logistics robots moving supplies, linen and waste. Inventory scanning robots. Autonomous mobile robots in manufacturing logistics. Common thread: mobility with a simple or no manipulation requirement, in semi-structured indoor space.
Deployed in narrower settings. Agricultural robots for specific crops and tasks. Hospitality delivery robots. Security patrol robots. Surgical systems, which are teleoperated rather than autonomous and are a separate field.
Working demonstrations, limited commercial deployment. General-purpose mobile manipulation, humanoid platforms performing varied tasks, robots operating alongside untrained people at close quarters doing physical work.
On humanoids specifically, the honest position is that investment and attention currently exceed deployment. The form has a real argument behind it, which is that human environments are built for human shape, and legs are seldom the hardest part. Manipulation and general capability are.
A workforce plan built on the third category is planning for a possibility. One built on the first is planning for a business that exists.
What the direction covers
The scope: locomotion and manipulation, social robots, and machines that assist at home and at work.
Four areas.
Mobility in human spaces. Navigation, obstacle avoidance, lifts and doors, and behaving predictably around people.
Manipulation. Grasping and handling varied objects, covered below.
Human interaction. How the robot signals intent, how people interpret it, and how it behaves when someone blocks it.
Deployment and operations. Commissioning into a live building and keeping a fleet running.
Where this sits in the domain
Humanoid and service robotics is the eighth of nine directions in Astra Trainer's robotics and autonomous systems domain, drawing on robotics engineering, control systems and computer vision, and sitting immediately before human-robot interaction and robot safety, which covers the shared-workspace question properly.
Partners deploying service robots typically need the operations layer more than the design layer, which is why this direction is often scoped with the advanced manufacturing domain's maintenance and asset management, and with supply chain and production operations for logistics settings. You can see the nine directions here.
Manipulation is the unsolved half
The clearest technical dividing line in this direction.
Moving a robot around a building is largely solved for practical purposes. Picking up arbitrary objects is not, and the reasons are worth understanding because they explain the deployment pattern.
Objects vary without limit. Shape, weight, rigidity, surface, fragility, how they are packed and how they are oriented.
Grasping requires judgement about physics. Where to grip, how hard, and what happens when you lift. Too little force and it slips, too much and it deforms or breaks.
Tactile feedback is limited. Humans adjust grip continuously from touch. Robot tactile sensing has improved and remains far behind.
Clutter makes it harder. Objects touching, overlapping and occluding each other, which is the normal state of a real container.
Failure is expensive. Dropping a parcel is inconvenient; dropping something fragile or hazardous is not.
This is why deployed service robots overwhelmingly move things rather than handle them, and why warehouse systems bring shelves to human pickers rather than picking. The division of labour reflects the technical boundary precisely.
Fleet operations is the business
The part that employs people and receives the least attention.
A deployed service robot fleet needs continuous operational support.
Uptime management. Robots get stuck, run out of charge, lose connectivity and encounter situations they cannot resolve. Someone monitors and responds.
Remote intervention. Many deployed systems rely on a remote operator who can take control or advise when the robot escalates. This is a real, currently existing job that is rarely described in the literature about the field.
Maintenance. Batteries, wheels, sensors, cleaning. Mundane and decisive for availability.
Site adaptation. Buildings change, and the robot's map and behaviour need updating.
Human factors on site. Training the staff who share the space, handling complaints, and managing the social reality of a machine in a workplace.
For an organisation deploying service robots, this is the capability to build, and it is almost always underestimated at purchase.
The roles, named
Robotics deployment and commissioning engineers. Getting a fleet working in a specific building.
Fleet operations specialists and remote operators.
Field service and maintenance technicians. The largest population as deployments grow.
Navigation and mapping engineers.
Manipulation researchers and engineers. Small, specialised, at the research boundary.
Human factors and interaction designers. How the robot behaves around people, which determines whether a deployment is accepted.
Safety engineers for machines operating around untrained people.
Customer success and site liaison roles. Unglamorous and frequently the difference between a renewed contract and a removed robot.
Who can be trained into it
Facilities and building services staff. Understand the buildings, the people in them and the operational rhythm. They are on site, they know why the corridor is blocked on Tuesdays, and they are rarely considered.
Logistics and warehouse operations staff. Know the material flow the robots are joining, which is what determines whether the deployment helps.
Field service engineers from any equipment industry. Already diagnose unfamiliar faults on customer sites under time pressure, which is exactly the fleet maintenance job.
IT support staff. Into fleet monitoring and remote operations, where the work resembles managing distributed systems.
Controls and automation technicians. Into commissioning and deeper diagnosis.
Healthcare logistics and portering staff, for hospital deployments, where knowing the actual workflow prevents the classic failure of a robot that technically works and disrupts care.
Machines operating around untrained people. Service robots working in spaces occupied by the public or by untrained staff are subject to machinery safety requirements and require a risk assessment covering foreseeable misuse, vulnerable users and emergency situations. Deployments in healthcare settings may carry additional requirements, and any robot performing a medical function is a regulated device. Training builds engineering and operational understanding. It does not constitute a site risk assessment or authorisation to deploy.
What to take from this
Everything a factory removes to make robotics work is what a service environment puts back, and public failure carries a different cost from a downtime statistic.
Mobility is largely solved and manipulation is not, which is precisely why deployed systems move things rather than handle them.
Humanoid form attracts attention beyond deployment, and legs are seldom the constraint. Manipulation is.
Fleet operations, remote intervention and maintenance are where the jobs are, and buyers consistently underestimate them.
And facilities, logistics and field service staff convert better than robotics graduates for most roles that actually exist.
Why is service robotics harder than industrial robotics?
Because the environment is not designed around the robot. People are untrained and unpredictable, layouts change, conditions vary, objects are arbitrary, and failures happen in public.
What is actually deployed commercially?
Warehouse mobile robots, floor cleaning, hospital logistics, inventory scanning and manufacturing logistics. The common thread is mobility with little or no manipulation, in semi-structured indoor space.
Why is manipulation still unsolved?
Objects vary without limit, grasping requires judgement about physics, tactile sensing remains far behind human capability, clutter compounds the problem, and failure can be expensive.
What should a buyer staff for?
Fleet operations: uptime monitoring, remote intervention, maintenance, site adaptation and the human factors of a machine in a workplace. This is consistently underestimated at purchase.
Who converts into these roles?
Facilities and building services staff, logistics operations staff, and field service engineers from any equipment industry, who already diagnose unfamiliar faults on customer sites.