Mobile manipulation is the branch of robotics in which a manipulator arm is mounted on a mobile base, so that the robot can both move through an environment and physically interact with objects in it. In manipulation, it removes the main limit of a fixed arm, the bounded robot workspace, by letting the base carry the arm to wherever the task is.
A mobile manipulator combines navigation, perception, grasp planning and motion planning in one system. The central technical challenge is coordinating base and arm as a single kinematic chain with redundant degrees of freedom, while coping with localization error, unstructured surroundings and the dynamic coupling between a moving platform and a loaded arm.
How a mobile manipulator is built
A mobile manipulator consists of three main subsystems: a mobile base, one or more arms with end effectors, and a sensor suite for localization and object perception. The base adds two or three planar degrees of freedom, and a typical six or seven joint arm brings the total to nine or ten degrees of freedom, more than the six needed to place a gripper in space.
The table below summarizes the usual base types and how each affects manipulation.
| Base type | Planar mobility | Effect on manipulation |
|---|---|---|
| Differential drive | Nonholonomic, cannot move sideways | Base repositioning needs multi-step maneuvers |
| Omnidirectional wheels | Holonomic, moves in any direction | Allows smooth whole-body motion |
| Legged platform | Handles stairs and rough ground | Body sway complicates arm accuracy |
| Rail or gantry carriage | One fixed axis | Extends reach along a known line |
| Humanoid torso on wheels | Usually holonomic | Two arms for bimanual tasks |
Sensors usually include lidar for navigation, depth cameras on the head or wrist for object pose estimation, joint encoders, and often a wrist force-torque sensor for contact tasks. Onboard batteries, compute and payload limits constrain arm size, so most mobile arms carry 3 to 15 kg, far less than stationary industrial robots.
Coordinating the base and the arm
Base and arm can be controlled sequentially or as one whole-body system. The choice determines how fast, smooth and accurate a task becomes, and it changes the planning and control software considerably.
| Strategy | How it works | Main trade-off |
|---|---|---|
| Move then manipulate | Base parks, then arm acts alone | Simple and robust, but slow |
| Base placement optimization | Planner picks a parking pose that maximizes reach | Better success, still sequential |
| Whole-body motion planning | Base and arm planned as one chain | Fluid motion, larger search space |
| Whole-body reactive control | Optimization or null-space control in real time | Handles disturbances, needs fast loops |
Whole-body approaches exploit kinematic redundancy. Inverse kinematics distributes motion between base and joints, and the extra freedom serves secondary goals such as avoiding joint limits, maximizing manipulability, or keeping the base still while the arm makes fine corrections. Frameworks such as MoveIt can treat the base as additional virtual joints so that collision checking and trajectory planning cover the entire robot.
Because the base is a less precise actuator than the arm, many controllers let the base follow a coarse trajectory while the arm compensates the residual error with visual servoing or force feedback. This division of labor exploits each subsystem where it performs best.
Mobile manipulation compared with related approaches
Mobile manipulation differs from fixed-base manipulation mainly in uncertainty and workspace size. A fixed cell knows its own position exactly, whereas a mobile robot must re-estimate its pose relative to every workstation it visits.
The comparison below shows where mobile manipulation fits among neighboring robot configurations.
| Configuration | Workspace | Typical position accuracy | Best suited for |
|---|---|---|---|
| Fixed industrial arm | Single cell | About 0.02-0.1 mm | High-volume repetitive tasks |
| Mobile manipulator | Whole facility | About 1-10 mm without local correction | Varied tasks at many stations |
| Autonomous mobile robot without arm | Whole facility | About 10-50 mm docking | Transporting goods only |
| Gantry robot | Large but fixed frame | About 0.05-0.5 mm | Large parts in one area |
The position values are indicative and depend heavily on calibration, localization method and payload. Mobile systems close the gap by referencing fiducial markers or workpiece features at each station, which can bring final accuracy close to that of a fixed arm.
Typical tasks and application domains
Mobile manipulation is chosen when tasks are spread across space and vary too much to justify dedicated fixed automation. Common examples include machine tending across several machines, order picking from shelves, laboratory sample handling, hospital logistics, household assistance and inspection with contact probes.
- Machine tending: one robot serves several CNC machines and moves between them as cycles finish.
- Warehouse picking: the robot drives to a shelf and performs pick and place or bin picking on mixed items.
- Laboratory automation: the robot opens instrument doors, loads plates and moves samples between stations.
- Domestic and service tasks: opening doors, loading dishwashers and fetching objects require dexterous manipulation in cluttered homes.
Door opening and drawer pulling are classic benchmarks because they need coordinated base motion under contact. The hand follows a constrained arc while the base moves, so impedance or force control is often essential to avoid excessive internal forces.
Limitations and common pitfalls
The biggest practical obstacle is accumulated uncertainty. Localization error, uneven floors, base compliance and dynamic sway all add to arm positioning error, so plans that work in simulation often fail on real floors.
| Problem | Root cause | Typical mitigation |
|---|---|---|
| Missed grasps after docking | Base pose error of several millimeters | Local re-detection of the object with wrist camera |
| Tipping risk with extended arm | Shifted center of mass and acceleration | Stability limits on payload, reach and speed |
| Slow task cycles | Strictly sequential base and arm motion | Whole-body planning and overlapping motions |
| Collisions in clutter | Outdated map or occluded obstacles | Continuous perception and online replanning |
| Short operating time | Arm and compute drain the battery | Opportunity charging and efficient scheduling |
Safety is also more complex than for a fenced cell. A mobile manipulator moves among people, so designers combine safety-rated laser scanners, speed and separation monitoring, and collaborative arm limits, and they treat the robot plus its arm as one combined risk assessment rather than two independent machines.
The simple explanation
Mobile manipulation means giving a robot arm a way to walk or drive. A normal factory arm is bolted to the floor, so it can only reach things within the length of its arm. Put that arm on wheels, and the robot can go to the object instead of waiting for the object to come to it.
It is similar to the difference between a person sitting at a desk and a person walking around a kitchen. The person in the kitchen can fetch, open and carry things anywhere, but must also watch where they step. The robot faces the same extra challenge: it has to know exactly where it stands before it reaches out.
How it is used in practice
A mid-sized machine shop runs four CNC lathes that each finish a part every 12 to 20 minutes. Instead of building a fixed robot cell for every lathe, the shop deploys one mobile manipulator with an omnidirectional base and a six-axis collaborative arm. The robot drives to whichever lathe signals that its cycle is complete, unloads the finished part and inserts a fresh blank.
Because the base parks with an error of a few millimeters, the robot first scans a fiducial marker on each machine with its wrist camera and corrects its arm trajectory, reaching a final placement accuracy of about 0.2 mm. The single robot keeps all four lathes running through an unattended night shift, and the shop reports higher machine utilization without dedicating floor space to four separate cells.