Home · Open data · Manipulation · Robot video

FMB: Functional Manipulation Benchmark

Roughly 31 hours across 22,550 real Franka Panda teleop trajectories of grasping, repositioning and insertion for assembly, 4 RGB-D views, force/torque.

Published by UC Berkeley RAIL, 2024-01

A Franka gripper inserting a peg beside a green shape board
Open data

Frame from the FMB: Functional Manipulation Benchmark dataset, UC Berkeley RAIL, CC BY 4.0.

Commercial use OK

CC BY 4.0

PriceFree

The publisher's licence permits commercial use and redistribution, with attribution.

Get dataset Opens functional-manipulation-benchmark.github.io in a new tab
From a terminal
wget https://rail.eecs.berkeley.edu/datasets/fmb/single_object_manipulation.zip

Four zip archives: Single-Object 545 GB, Assembly 1 86 GB, Assembly 2 77 GB, Assembly 3 70 GB. Single-request downloads from a university server; no token. Also included in Open X-Embodiment as fmb (RLDS) but that mirror was not verified here.

Specification, as published

Hours (derived)
31 h
Episodes (stated)
22,550 episodes
Scale
We collected a total of 22,550 trajectories across two different tasks on a Franka Panda arm.
Task
Assembly, Pick & place
Environment
Indoor – lab
Modality
RGB, Depth, Proprioception, Actions, Force/torque
Capture
CCTV Scene
In frame
Robot hands
Captured in
North America
Embodiment
Franka Panda
Resolution
256x256
Formats
NPZ/NumPy, Other
Size
545 GB (Single-Object) + 86 GB + 77 GB + 70 GB (Assembly 1 to 3)
Hosted on
rail.eecs.berkeley.edu

Figures are the publisher's own. Kinetic Blocks has not measured this dataset. Estimated from the paper: 22,550 trajectories of about 5 seconds each. The publisher states no total.

About this dataset

FMB is a real-robot benchmark from UC Berkeley focused on the sub-skills of assembly: grasping, repositioning and inserting objects into boards. A Franka Panda arm was teleoperated with a SpaceMouse over 54 single objects and three sets of interlocking multi-object assemblies. Each trajectory carries two global and two wrist RGB-D views plus end-effector pose, velocity, force/torque, joint state and gripper state, stored as NumPy files in four large zip archives.

Publisher's own task names: grasping, repositioning (place, regrasp, rotate), insertion, single-object multi-stage assembly, multi-object assembly.

What Kinetic Blocks did, and did not do

Indexed and linked from the publisher. Not hosted, verified or graded by Kinetic Blocks. The publisher's terms govern. The licence shown is the one stated on the publisher's page, linked above. Nothing on this page is a claim by Kinetic Blocks.

Want the verified ones too?

Request access to the marketplace, where every paid listing is verified against its files and graded before it goes on sale.