ARMBENCH DATASET

The Stow Success Prediction dataset contains 72,000+ complete stow cycles with data collected from Amazon Vulcan Stow workcells located in an Amazon warehouse. The dataset supports the Success Prediction task: Predicting the binary outcome of a stow attempt (success or failure) based on the provided images and tabular data. The dataset is organized as follows:

Images: All images are named after their corresponding run_id for cross-referencing:
- Induct: Items on the conveyor before stowing
- Pre-stow Pedestal: Bin before stowing
- Mid-stow Pedestal: Bin during the stow process
- Post-stow Pedestal: Bin after stowing
The pedestal images are captured from fixed cameras, showing the specific bin being stowed into at different stages of the stow process.

Tabular Data: A comprehensive table with run_id as the primary identifier and all other feature columns as described below.

Category	Feature Name	Description	Data Type
Metadata	run_id	Unique identifier for each stow cycle	string
Outcome Features	computed_success	Boolean indicating if the stow attempt was a success (True) or failure (False)	bool
	computed_unproductive	Boolean indicating if the stow attempt was unproductive (True) or not (False)	bool
	computed_defect	Boolean indicating if the stow attempt was a defect (True) or not (False)	bool
Item Features	itemLengthPerception	Item length estimate produced by perception [m]	float
	itemWidthPerception	Item width estimate produced by perception [m]	float
	itemHeightPerception	Item height estimate produced by perception [m]	float
	itemLengthManifest	Item length estimate from the manifest [m]	float
	itemWidthManifest	Item width estimate from the manifest [m]	float
	itemHeightManifest	Item height estimate from the manifest [m]	float
	weightKg	Weight of item in KG (from manifest)	float
	itemPackagePackagingType	String indicating the item packaging type (box, polybag, etc.)	string
	itemPackageShape	String indicating the item packaging shape	string
	isFragile	Boolean for if the item is fragile	bool
	isFoldable	Boolean for if the item is foldable	bool
	containsLiquidClassification	Boolean for whether the item contains liquid	bool
	bindingText	String indicating the use category of the item (e.g., 'Kitchen', 'Office Product', 'Apparel')	string
Bin Features	binDepth	Bin depth measurement [m]	float
	binWidth	Bin width measurement [m]	float
	binHeight	Bin height measurement [m]	float
	binVolume	Bin volume measurement [m³]	float
	binGauPerception	Perception estimate of the gross area utilization [m²]	float
	gluScore	Metric for gross linear utilization	float
	gcuScore	Metric for gross cubic utilization	float
	numItems	Number of items in the bin prior to the current stow attempt	int
	numStows	Number of stow attempts to this bin prior to the current attempt	int
	numSuccessfulStows	Number of successful stows to the bin we are attempting to stow to	int
	isEdge	Boolean indicating if the bin we are attempting is on the edge of the pod or not	bool
	podType	Categorical feature indicating the type of pod being stowed to	string
Stow Strategy Features	stowStrategy.type	String indicating the stow strategy type (e.g., SLOT_WEDGE)	string
	stowStrategy.direction	Blade sweep direction ('MINUS_Y', 'PLUS_Y')	string
	stowStrategy.slotEdgeType	String for slot edge type ('NO_EDGE', 'RIGHT', 'LEFT')	string
	ssBinSweep	Boolean indicating if stow attempt was a bin sweep	bool
	ssSlotWedge	Boolean indicating if stow attempt was a slot wedge	bool
	ssStacking	Boolean indicating if stow attempt was stacking (rotated wedge)	bool
	slotWidth	Width of the slot we are stowing to in this cycle [m]	float
	maxL3SlotWidth	Maximum slot width of all BinSweep affordances (L3 type slots)	float
Space Metrics	mlo	Missed linear opportunity metric	float
	dlo	Detected linear opportunity metric	float
	dao	Detected area opportunity metric	float
	dco	Detected volume opportunity metric	float
	dloSweep	Detected linear opportunity metric for Bin Sweep (L3) affordances	float