Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation

A research team has released Humanoid Everyday, a large-scale dataset designed to train humanoid robots for open-world manipulation tasks, alongside a cloud-based evaluation platform for standardized benchmarking [1]. The dataset aggregates 10.3k trajectories and over 3 million frames of multimodal sensory data, spanning 260 distinct tasks organized into seven broad categories [1][2]. The data includes RGB, depth, LiDAR, and tactile inputs, all accompanied by natural language annotations [1][2]. The work was submitted to the arXiv preprint repository in October 2025 and revised in June 2026 [1]. The researchers note that most existing robot-learning benchmarks focus on stationary arms, while the few available humanoid datasets are often limited in task diversity or confined to fixed environments [1][2]. Humanoid Everyday addresses these gaps by incorporating dextrous object manipulation, human-humanoid interaction, and locomotion-integrated actions [1][2]. The data was collected through a human-supervised teleoperation pipeline [1][2]. To facilitate consistent comparisons, the team introduced a cloud-based evaluation platform where researchers can deploy their policies in a controlled setting and receive performance feedback [1][2]. The platform is intended to provide a standardized method for assessing learning-based policies on humanoid data, an area where few benchmarks currently exist [1][2]. arXiv, where the paper is hosted, is an open-access repository of electronic preprints that has been operating since August 1991 and now receives about 24,000 submissions per month [8]. The platform also supports community-developed tools through arXivLabs, a framework that allows collaborators to build features such as bibliographic explorers and code finders directly on article pages [6][7]. The release of Humanoid Everyday comes as humanoid robots are demonstrating increasingly complex full-body capabilities, from locomotion to manipulation [1][2]. The dataset, data-collection code, and evaluation website have been made publicly available on the project’s website [1][2].