Short answer
Trailer loading and load-building data records people stacking cartons or pallets into a trailer or container so the load is stable, fits and unloads in the right order. Distributors and 3PLs hold dock video and load plans or manifests. It can support stacking-sequence understanding and perception of partial loads, but not robot actions, and it is separate from unloading data.
The deployment problem
Loading is a planning problem that happens at the pace of a dock. A loader decides where each carton goes, how it is oriented, what supports it and what it blocks. The aim is a load that uses the space, does not shift in transit and can be unloaded at each stop. Some freight is floor-loaded, carton by carton; some is palletized.
Unloading, covered in trailer and container unloading, starts from a finished stack and takes it apart. Loading is the reverse in direction but not in difficulty: the decision about where a carton goes is yours, and the wrong decision shows up only later as a collapsed wall or an unusable cube. That is why a separate page is warranted. Palletized stacking is a related task, covered in case picking and mixed palletizing, and dock-to-dock flows in cross-docking tasks.
Task anatomy
| Step | Decision or action | Failure modes |
|---|---|---|
| Read the load plan | Stop sequence, weights, fragility | Plan not followed |
| Stage freight | Bring cartons or pallets to the door | Staging blocks the dock |
| Choose the next item | Pick by size, weight and destination | Heavy item on light one |
| Place | Set carton against the wall or the previous layer | Gaps, overhang |
| Build the wall | Layers across trailer width | Leaning wall, crushed bottom layer |
| Secure | Dunnage, straps, load bars | Missing restraint |
| Verify | Count, scan, photograph | Count mismatch |
| Close out | Doc, seal | Missing record |
What this data can support, and what it cannot
Can support. Recognition of partial-load states; estimation of free space; detection of placement gaps and lean; step segmentation; links between a load plan and what was actually loaded; and evaluation of stack stability reasoning when outcomes are logged. Manifests and load plans give structure: unit identifiers, destinations and sequence. The GS1 Serial Shipping Container Code is an 18-digit key that identifies a logistic unit such as a case or pallet and keys despatch advice data (GS1); where a site uses it, a loaded unit can be linked to its record.
Cannot support. Existing dock video rarely shows why a loader chose a position. It does not include robot joint states, commanded actions or the forces needed to push a carton into place. A person lifting from a pallet jack and a humanoid with a different reach and payload face different constraints; see the embodiment gap and what human demonstration video can and cannot do. Video also does not tell whether the load arrived intact; that outcome comes from claims or delivery records. And the data covers one dock's loading style, not general packing.
Lifting risk matters here. NIOSH's Revised Lifting Equation estimates the risk of low back disorders associated with manual lifting tasks (NIOSH, 2021). Loaders' postures in video reflect those hazards; a robot team should not treat human lifting posture as a motion target.
Existing footage and commissioned recording
Existing assets: dock cameras, door cameras, load-plan or manifest exports, scan logs and trailer-close photos. They show real pace and real freight mix but usually at a wide angle and with the trailer interior hidden. Commissioned recording can add a camera inside the trailer or on the loader, depth sensors, and a predefined set of freight; it requires site access, consent and time. The comparison is set out in existing footage vs commissioned recording. If your target needs views inside the trailer, discuss feasibility with SourceX rather than assuming existing footage has them.
Illustrative record set
# Illustrative only - synthetic values
load_id: load_00981
site_id: site_036
trailer: 53ft_dry_van (type stated)
load_plan: {stops: 3, units: [{unit_id: u_01, dest_stop: 1, weight_kg_class: medium}]}
video:
dock_wide: fixed, door-facing
interior: optional, depth camera
events: [unit_staged, unit_placed, wall_complete, restraint_applied, trailer_sealed]
labels_added: [placement_position, layer_index, stability_flag, gap_mm_class]
outcome: [delivery_damage_flag if available]
excluded: [driver_identity, consignee_names, robot_state]
holdout: by site and by freight type
Quality and acceptance pointers
- Interior visibility. Check how much of the trailer interior the camera can see once the wall rises.
- Plan-to-video match. Verify that units in the plan can be found in the footage.
- Freight diversity. Carton sizes, stiffness, pallets vs floor-load, mixed vs uniform.
- Outcome data. Damage or shift claims link a stacking decision to a result.
- Timestamps. Door open, first unit and seal times should be consistent with video.
- Splits. Hold out whole docks and freight types.
Supplier types
Distributors, 3PLs and retailers' outbound docks. Freight belongs to shippers or consignees, and drivers visit the dock; both appear in video. Read client-owned inventory rights, 3PL providers as warehouse data suppliers and pallet inspection and grading for related records.
Limitations and additional preparation
Load-building data needs labels for placement, layer and stability that annotators must add. Camera views are likely to be insufficient for 3D reconstruction of the stack. Records are often paper or PDF and need digitizing. Weight and dimension values may be nominal; measured values come from dimensioning and weighing station records. A load-building policy for a humanoid also needs robot-specific data for lifting, carrying and placing at trailer scale, recorded on the target platform.
Reading a sample of loading footage
Ask for a load from start to close, not highlights. Watch for four things. First, can you see the placement decision, meaning the moment the loader chooses a position, or only the placement? Second, does the footage show the load being checked against the plan? Third, how much of each load is visible once the wall is higher than the camera? Fourth, are failures present: a leaning wall that is rebuilt, a carton that is moved twice, a stop where freight is removed and reloaded.
If most loads are uniform cartons on pallets, the data says little about floor-loaded mixed freight. Match the sample to your target: palletized freight needs different perception and different grasps than loose cartons. Note that the footage may also show vehicles, drivers and other people outside the company, which affects consent and redaction.
Questions to ask suppliers
- Are loads floor-loaded, palletized or both, and in what proportion?
- Which cameras show the trailer interior, and from where?
- Are load plans or manifests available, and how do they map to units in video?
- Do records include damage, shift or claims outcomes?
- What range of freight (size, weight, stiffness) appears?
- How many docks and trailer types are included?
- Are drivers, loaders or consignee names visible in video or documents?
- Who owns the freight records, and have shippers approved licensing?
- Are load-bar or strap events logged?
Discuss your requirements
If you can describe freight types, camera views and plan fields, describe the requirement to SourceX. SourceX reviews it, follows up about potential sources among US businesses, checks the data and the supplier's licensing permissions, and coordinates the license and delivery; commissioned recording would be scoped case by case. A request does not guarantee a match. Start at the humanoid training data guide or the warehouse and fulfillment hub; catalog context is on the supply chain and logistics data page.
Discuss your humanoid training-data requirements with SourceX
Sources
- Serial Shipping Container Code (SSCC) — logistic unit key and link to despatch data.
- Applications Manual for the Revised NIOSH Lifting Equation (DHHS (NIOSH) Publication No. 94-110) — risk estimate for manual lifting tasks.