Environments & Tasks

MIKASA-Robo-VLA contains 90 tasks across 10 memory types and three horizon splits.

Split	Tasks	Horizon (steps)	Typical episode length (at 25 Hz)
Short	38	25 – 200	1 – 8 s
Medium	30	201 – 600	8 – 24 s
Long	22	601 – 2160	24 – 86 s

Memory Types

The table below gives an operational definition of each memory type: what information must persist across time, why current-frame observations are insufficient, and what kind of history the agent must use.

Memory Type	# Tasks	What must be remembered or tracked	Why a reactive policy fails
Object	18	A previously observed object attribute: colour, shape, identity, or a shape-colour binding.	The target attribute disappears before the agent must choose.
Spatial	14	A hidden placement, earlier reference pose, motion context, or spatial relation needed for a later manipulation decision.	The current scene may omit the hidden target, earlier reference state, or trajectory context needed for the spatial action.
Capacity	12	Multiple unordered items or object attributes, or a count that summarises cues exposed together or one at a time.	Remembering only one item is insufficient; the agent must preserve the full set or its task-relevant aggregate.
Temporal	12	Information accumulated over time, such as blink counts, elapsed-step counts, timed cues, or stage-dependent events.	The decision depends on the history of observations, not on a single visible cue.
Negative	9	Which candidates were present before, so the agent can identify the missing or odd-one-out candidate later.	The correct answer is defined by exclusion rather than by matching a visible cue.
Sequential	6	The order in which multiple targets or cues appeared and must later be executed.	The same target set can require different actions when its remembered order changes.
Procedural	6	A path-like motor procedure or sequence of movements that must be reproduced during execution.	The current state does not encode the remaining procedure; the agent must recall how the motion should continue.
Prospective	5	A delayed intention that must remain active while another task is being completed.	The early goal is no longer salient when the final action is required.
Tracking	4	A belief about hidden object identity or position while objects move or exchange roles.	The correct state changes during the episode and must be updated continuously.
Checklist	4	Which required conditions or items have already been tested, verified, or completed, and what result each check produced.	The current frame cannot recover the completed checks, their outcomes, and the remaining required items.

See Core Concepts for a description of each memory type, and Benchmarking for the canonical evaluation protocol.

This section is the task catalog. Each row links to one environment-family page, where all difficulty settings implemented by the same Python file are documented together. For example, Remember Color covers RememberColor3-VLA-v0, RememberColor5-VLA-v0, RememberColor9-VLA-v0, and their long-horizon variants: RememberColor3-Long-VLA-v0, RememberColor5-Long-VLA-v0, RememberColor9-Long-VLA-v0.

Use the table for quick task selection, then open a task page for mechanics, wrapper recommendations, render previews, and collection commands.

Task Overview

← scroll →

Preview	Task	Episode Length	Horizon Split	Memory Type	Difficulty	Long Variant	Data Source	Language Instruction
	ShellGameTouch-VLA-v0	short 30	Short	Spatial	single setting		PPO	Observe which cup hides the ball, wait, then touch that cup.
	ShellGamePush-VLA-v0	short 30	Short	Spatial	single setting		PPO	Observe which cup hides the ball, wait, then push that cup forward.
	ShellGameShuffleTouch-VLA-v0	short 60; medium 600	Short / Medium	Tracking	single setting	yes (600 steps)	PPO / MP	Observe which cup hides the ball, track the cups as they shuffle, then touch the correct cup.
	ShellGameColorLampTouch-VLA-v0	short 30	Short	Spatial	single setting		PPO	Observe which color is under each cup, then touch the cup matching the lamp color.
	ShellGameShuffleColorLampTouch-VLA-v0	short 60; medium 600	Short / Medium	Tracking	single setting	yes (600 steps)	PPO / MP	Observe which color is under each cup, track the cups as they shuffle, then touch the cup matching the lamp color.
	InterceptSlow/Medium/Fast-VLA-v0	short 60	Short	Spatial	slow / medium / fast		PPO	Intercept the rolling ball by moving to its path and deflecting it toward the target.
	InterceptGrabSlow/Medium/Fast-VLA-v0	short 60	Short	Spatial	slow / medium / fast		PPO	Intercept the rolling ball and grasp it to stop it.
	RotateLenientPos/PosNeg-VLA-v0	short 60	Short	Spatial	positive / signed angles		PPO	Rotate the peg by {angle_deg} degrees to match the target angle.
	RotateStrictPos/PosNeg-VLA-v0	short 90	Short	Spatial	positive / signed angles		PPO	Rotate the peg by {angle_deg} degrees to match the target angle while keeping the center of the peg in place.
	TakeItBack-VLA-v0	short 60	Short	Spatial	single setting		PPO	Push the cube onto the red target, and when the target changes color, return the cube to its original position.
	RememberColor3/5/9-VLA-v0	short 25; medium 600	Short / Medium	Object	3 / 5 / 9	yes (600 steps)	PPO / MP	Observe the cube’s color, wait, then touch the cube of the same color.
	RememberShape3/5/9-VLA-v0	short 25; medium 600	Short / Medium	Object	3 / 5 / 9	yes (600 steps)	PPO / MP	Observe the object’s shape, wait, then touch the object of the same shape.
	RememberShapeAndColor3x2/3x3/5x3-VLA-v0	short 25; medium 600	Short / Medium	Object	3x2 / 3x3 / 5x3	yes (600 steps)	PPO / MP	Observe the object’s shape and color, wait, then touch the object of the same shape and color.
	FindImposterColor3/5/9-VLA-v0	short 25	Short	Negative	3 / 5 / 9		PPO	Observe the cubes shown, wait, then touch the cube whose color was not present before.
	FindImposterShape3/5/9-VLA-v0	short 25	Short	Negative	3 / 5 / 9		PPO	Observe the shapes shown, wait, then touch the object whose shape was not present before.
	FindImposterShapeAndColor3x2/3x3/5x3-VLA-v0	short 25	Short	Negative	3x2 / 3x3 / 5x3		PPO	Observe the objects shown, wait, then touch the object whose shape and color combination was not present before.
	BunchOfColors3/5/7-VLA-v0	medium 400; long 700	Medium / Long	Capacity	3 / 5 / 7	yes (700 steps)	MP	Observe which colored cubes appear during the cue, wait, then touch all of them in any order and press the center button.
	SeqOfColors3/5/7-VLA-v0	medium 400; long 800; long 1000; long 1200	Medium / Long	Capacity	3 / 5 / 7	yes (800 / 1000 / 1200 steps)	MP	Observe which colored cubes appear during the cue, wait, then touch all of them in any order and press the center button.
	ChainOfColors3/5/7-VLA-v0	medium 400; long 800; long 1000; long 1200	Medium / Long	Sequential	3 / 5 / 7	yes (800 / 1000 / 1200 steps)	MP	Observe which colored cubes appear during the cue, wait, then touch all of them in the same order as the cubes were shown and press the center button.
	TraceShapeEasy/Medium/Hard-VLA-v0	medium 250; medium 300; medium 350	Medium	Procedural	easy / medium / hard		MP	Watch the red cube trace a shape on the table. When the lamp turns green, pick up the green cube and trace exactly the same shape.
	TraceShapeSeqEasy/Medium/Hard-VLA-v0	long 1500	Long	Procedural	easy / medium / hard	yes (1500 steps)	MP	Watch the red cube trace a sequence of shapes. When the lamp turns green, pick up the green cube and trace the same sequence in order. After finishing all shapes, press the button to submit your answer.
	BlinkCountButtonPressEasy/Medium/Hard-VLA-v0	short 150; short 200; medium 300; long 1200	Short / Medium / Long	Temporal	easy / medium / hard	yes (1200 steps)	MP	Count how many times the blue lamp blinks, press the red button exactly that many times when the red lamp turns green, then press the black button to submit your answer.
	TimedTransferEasy/Medium/Hard-VLA-v0	short 200; medium 250; medium 300; medium 600; long 900; long 1200	Short / Medium / Long	Temporal	easy / medium / hard	yes (600 / 900 / 1200 steps)	MP	When the white lamp turns green, start counting steps from that exact moment. Move the blue cube from the green disc to the red disc exactly on step 100 of that count.
	BatteriesCheckerEasy-3/6-VLA-v0	medium 540; long 1080	Medium / Long	Checklist	3 / 6 batteries	yes (1080 steps)	MP	Find all working batteries by inserting each one into the socket, observing the lamp result, and then pressing the button to confirm.
	BatteriesCheckerHard-3/6-VLA-v0	long 1080; long 2160	Long	Checklist	3 / 6 batteries	yes (1080 / 2160 steps)	MP	Find all working batteries by inserting each one into the socket, observing the lamp result, returning it from the socket to its initial slot, and then pressing the button to confirm.
	GatherAndRecall1/3/5/7/9-VLA-v0	short 200; medium 400; medium 600; long 800; long 1000	Short / Medium / Long	Prospective	1 / 3 / 5 / 7 / 9	yes (800 / 1000 steps)	MP	Move all cubes onto the disc. A lamp will briefly flash while you work. After all cubes are placed, press the button matching the flash color.