A 90-day playbook for your first mobile manipulation task

Why the first task is the most important decision

The first mobile manipulation task in a facility sets the template for everything that follows. It calibrates the plant's integrators, safety engineers, and operators. It surfaces the site-specific quirks of the floor environment, the navigation system, and the arm-base integration. It produces the first real data on productive utilisation and downtime frequency.

A well-chosen first task makes the second deployment easier. A poorly chosen first task — one that is too complex, too precise, or too poorly structured — produces months of debugging, erodes internal confidence in the technology, and sometimes kills the programme entirely.

The 90-day structure below is designed to maximise the probability that the first task reaches stable, measured production operation within a single quarter. It is intentionally conservative: the goal is a single working task producing reliable data, not a showcase of the platform's full capability.

Days 1–14: Task selection and site assessment

The ideal first task

The ideal first task for a mobile manipulation deployment has the following characteristics:

Moderate tolerance requirement (±1–3 mm acceptable). A task where fiducial-anchored re-localisation can reliably achieve the required accuracy without exotic docking infrastructure. Loading labelled vials into a centrifuge, loading blanks into a lathe with a broad chuck range, or placing parts into fixtures with generous clearances all qualify.

Two to four stations, arranged within 40 metres. A small, compact circuit produces fast iteration cycles. If the re-localisation calibration at station 2 is wrong, you discover it in minutes, not hours.

No operators in the manipulation work envelope during the arm's operation phase. The first task should be designed so the arm operates in a zone that is physically clear of humans while the arm is moving. This simplifies the safety case and the risk assessment considerably. Collaborative operation (arm working alongside humans) is a valid long-term goal but the wrong starting point.

Consistent part geometry and orientation. Bin picking from randomly oriented parts is a separate discipline requiring advanced 3D vision and grasp planning. For the first task, present parts consistently — in a tray, magazine, or nest — so the arm's pick programme is deterministic.

A task currently performed by an operator on a repeatable schedule. Not a task that requires judgement, not a task performed irregularly. A consistent task gives you a baseline to measure against.

Site assessment checklist

Before hardware delivery, the site must be assessed on these dimensions:

• Floor map drafted. Complete floor-plan with all permanent obstacles, doorways, aisle widths, and the proposed station positions. Include overhead dimensions if arm raise creates clearance concerns.
• Floor surface evaluated. Any surface transitions (threshold strips, drain covers, floor paint edges) documented. Mobile bases tolerate ±10–15 mm surface irregularities; larger transitions need remediation.
• WiFi or wired connectivity confirmed at all station positions. Fleet management software requires reliable connectivity. Conduct a signal survey before mapping.
• Power outlets located for charging dock placement. Charging dock requires a standard industrial outlet within 1–2 metres of the dock position.
• Lighting conditions measured. Fiducial-based re-localisation vision systems work within a specified lux range. Measure ambient light at each proposed station position; identify whether supplemental lighting is needed.
• Safety perimeter space identified. The robot needs a safety-scanner coverage zone around the base. Identify whether existing aisles provide adequate clearance or whether permanent obstacles need to be relocated.

Days 15–30: Fixturing and perception design

Why fixturing is the highest-leverage investment

The single investment that most improves mobile manipulation reliability is a well-engineered fixture at each station. A fixture that presents parts in a consistent, repeatable orientation eliminates the need for complex perception and reduces re-localisation error from a critical constraint to a manageable one.

A good first-task fixture has these properties:

• Kinematic locating features. Two-point contact (pin + slot, or cone + flat) that repeatably positions the part to within ±0.5 mm without requiring the operator to "seat" it carefully. Kinematic mounts from the precision machining world are overkill but instructive as a design reference.
• Visual contrast for fiducial detection. If using printed fiducials (AprilTags, ArUco markers), the mounting surface should provide high contrast. A light-coloured marker on a dark machined surface performs more reliably than a marker on a reflective chrome fixture.
• Defined approach clearances. The fixture approach geometry — the path the gripper takes to reach the part — should have at least 20–30 mm clearance on all sides at the programmed approach. This accommodates navigation error without collision.

Fiducial placement rules

For fiducial-anchored re-localisation:

1. One fiducial per station, rigidly mounted. The fiducial must be mechanically fixed to the fixture or the machine frame — not taped to a cart that operators move. Any fiducial that can be casually repositioned will be, and the re-localisation will fail silently.

2. Fiducial in the arm's camera field of view at the approach pose. The re-localisation sequence typically has the arm move to a pre-defined "detection pose" near the station, detect the fiducial, compute the correction transform, then proceed to the pick. The detection pose must be reachable from the expected base parking range, and the fiducial must be visible from that pose.

3. Fiducial size calibrated to detection range. A 100 mm × 100 mm AprilTag detected at 400–600 mm range provides adequate pixel density for sub-millimetre correction. Smaller tags at the same range reduce accuracy.

4. Fiducial occluded when not in use. Operators who casually place objects in front of the fiducial cause silent re-localisation failures. A recessed mounting position that is not in the natural operator work area minimises accidental occlusion.

Days 31–60: Safety commissioning

The dual safety obligation

As described in The real cost of a mobile manipulation cell, a mobile manipulator operates under two simultaneous safety regimes: ISO 3691-4 for the mobile platform and ISO/TS 15066 for the collaborative arm. Both must be addressed in the commissioning safety documentation.

Before any live testing begins, the following must be complete:

• Risk assessment documented (ISO 12100). For each identified hazard: the harm, severity (S1 minor/S2 serious), exposure frequency (F1 rarely/F2 frequently), possibility of avoidance (P1 possible/P2 barely possible), and the required Performance Level or Safety Integrity Level of the safeguard.
• Mobile platform zones configured and validated. Most mobile bases use safety-rated laser scanners with configurable protection zones: a hard stop zone (robot halts if entered) and a warning zone (robot decelerates). These must be physically measured and validated by triggering each zone with a test object — not just software-confirmed.
• Arm operating modes defined. The arm should have at least three operating modes: transit (arm folded, power limited), approach (arm moving in free space, reduced speed), and task (arm in manipulation sequence, full capability within cobot safety limits). Each mode transition must be governed by a hardened state machine, not ad hoc code.
• Emergency stop function tested end-to-end. Press the physical E-stop, verify that both the base drive and the arm halt within the required time. Verify that the E-stop is not reset by software alone — it must require a physical key or deliberate two-step action to clear.
• Restart interlock confirmed. After any E-stop or protective stop, the system must not resume motion without a deliberate operator action (not automatic restart). This is an ISO 3691-4 requirement for AGVs and a functional safety requirement for cobots.

Cobot-specific safety validation

For the collaborative arm operating in shared workspace:

• Power and force limiting (PFL) validated. Most cobots implement PFL as a standard safety function. Validate with a biomechanical force gauge (or approved equivalent method) that the arm's contact forces at the programmed speeds are within ISO/TS 15066 limits for the body regions that could be contacted.
• Speed and separation monitoring (SSM) if applicable. If operators are present during arm operation (not the first-task recommendation, but plan for it), SSM reduces arm speed as operators approach. Validate the detection range and speed reduction response.
• Tool and payload verification. The safety controller's payload settings must match the actual gripper and part weight. An underreported payload causes the safety system to miscalculate contact forces.

Days 45–75: Integration and programming

Navigation mapping

The navigation map should be built with the floor in its operational state — with typical obstacle distributions, operators at their workstations, and any regularly-used equipment in place. A map built on an empty floor at midnight will generate excessive replanning events during production shifts.

Mark the navigation map with:

• No-go zones (areas the robot must never enter)
• Speed-limited zones (near pedestrian crossings or blind corners)
• Preferred lanes (to reduce conflict with forklift paths)

Map updates should be a defined procedure, not an ad hoc operation. A floor layout change that is not reflected in the map will cause navigation failures.

Station programming sequence

For each station, the programming sequence is:

1. Drive base to parking position manually. Record the target pose.
2. Verify safety zone function at this position. Confirm no fixtures or machine structures intrude into the protection zones.
3. Perform detection pose approach. Move arm to detection pose; verify fiducial is visible and the detection algorithm returns a stable result over 10 trials.
4. Calibrate re-localisation transform. Use the fiducial to compute the correction transform; validate by commanding a test pick and measuring end-effector accuracy with a test pin.
5. Programme full task path. Programme pick, transport, place, and return sequence with approach and retreat waypoints.
6. Run 50 consecutive trials unattended. Log every success, miss, fault, and recovery. A task that fails more than 5 percent of trials is not ready for production.

Days 75–90: Measurement and handoff

The five KPIs that matter

At the end of the pilot, report on five numbers — not vendor-selected metrics, but the numbers that reflect production reality:

If any KPI falls below target, diagnose before declaring the pilot a success. Common failure patterns:

• Task success <95%: usually a fixturing tolerance or re-localisation calibration issue. Re-examine fixture design.
• Utilisation <50%: usually a circuit design problem — stations too far apart or station cycle times too short. Redesign circuit or add stations.
• Recovery events more frequent than every 8 hours: usually a navigation obstacle or fiducial occlusion pattern that occurs at shift change or during breaks. Shadow a shift to observe.

Handoff documentation

For every deployment that transitions from pilot to production, produce:

• Navigation map with annotated no-go zones and preferred lanes
• Per-station: fixture drawing, fiducial position, detection pose, re-localisation calibration record, task programme
• Risk assessment document and safety validation records
• Operator procedure for manual takeover (what to do when the robot stops unexpectedly)
• Maintenance schedule (battery check, scanner cleaning, fiducial inspection, joint lubrication per arm manufacturer specification)

A mobile manipulation deployment without a maintenance schedule drifts toward increasing downtime over months — and the degradation is gradual enough that it often goes unnoticed until a major fault occurs.

What to read next

Choosing and deploying a platform is one challenge; selecting the right vendor in the first place is another. See The mobile manipulator vendor RFP for the questions to ask before signing a purchase order — including the red flags that separate genuinely production-ready platforms from those that are still research-grade.