Record Training Data

Record robot training data with Alia360 3D

We run capture programs with Alia360 3D recording the room and a first-person camera recording the worker. Every clip on this page was recorded in a working environment. Customer names are withheld by agreement.

Write to sales@dreamvu.ai about robot data.

( 01 )

Vision-language data

Robot training data

Multi-view video of real scenes from the room and from the worker's view. Built for adding questions and answers about what is happening in the scene. Use it to train and test spatial reasoning, physical reasoning and embodied action.

CaptureEgo + exo, synchronized

AnnotationCapability-probed Q&A pairs

FormatVideo with paired Q&A. Schema agreed per program.

A sample of the VLM library. The PRISM-100K open subset is on Hugging Face. Full set under NDA.

( 02 )

Navigation data

Robot training data

Continuous walks through working spaces from the worker's point of view. Trajectories, free space, and obstacle events annotated throughout. Includes 6-DoF pose tracking.

CaptureHead-mounted ego camera, paired with Alia360 3D

AnnotationTrajectory, free space, obstacle events

FormatVideo with 6-DoF pose track. Full dataset under NDA.

A sample of the navigation library. Full dataset available under NDA.

( 03 )

Loco-manipulation data

Robot training data

Work that involves moving through the space, not just working at a station. Approach, reach, whole-body positioning and two-handed work at close range, captured from the room, the worker's view and the wrist at the same time.

CaptureEgo + exo + wrist, synchronized

AnnotationHand pose, contact, action boundaries, task breakdown

FormatLeRobot (RLDS), Open X-Embodiment

Stocking and replenishment

Grocery · ego + exo + wrist

Pick and pack

Retail store · ego + exo + wrist

Food prep, knife work

Commercial kitchen · ego + exo + wrist

A sample of the loco-manipulation library. Full dataset available under NDA.

How it works

Record, sync, process, export

Record

Alia360 3D records the room. A first-person camera records the worker. Wrist cameras are added for close manipulation. Your own cameras can be included.

Sync

All streams are aligned frame for frame on the compute unit.

Process

Upload recordings to the DreamVu processing pipeline. It labels them: segmentation, tracking, hand pose, contact, action boundaries and task breakdown. You pay only for what you process.

Export

Download in LeRobot, RLDS or Open X-Embodiment format, or a custom schema.

Quality

Every capture passes four checks before delivery

Calibration

Checked every session.

Alignment

All camera streams confirmed in sync before annotation starts.

Review

A person reviews every batch.

Rejection

Batches below threshold are captured again.

Published results

Tested on public models

We test recordings made with our cameras on public models and publish the results.

arXiv · July 2026RetailSMV: Exocentric vs. Egocentric Adaptation of Foundation Video World Models in RetailWhich camera viewpoint helps video world models learn.A. Rouhi, R. Aggarwal, P. Sakurikar, A. M. Namboodiri, S. P. Reddi

Embodied AI Workshop, CVPR · June 2026PRISM-World: Teaching Video World Models the Physics of Retail EnvironmentsTeaching video world models how real environments behave.A. Rouhi, R. Aggarwal, P. Sakurikar, A. M. Namboodiri, S. P. Reddi

arXiv · May 2026SABER: A Scalable Action-Based Embodied Dataset for Real-World VLA AdaptationAn action dataset for vision-language-action models.N. Menga, P. Sakurikar, A. Rouhi, R. Aggarwal, A. M. Namboodiri, S. P. Reddi

arXiv · March 2026PRISM: A Multi-View Multi-Capability Retail Video Dataset for Embodied Vision-Language ModelsA multi-view video dataset for vision-language models.A. Rouhi, P. Sakurikar, S. S. Reddy, N. Menga, A. Govil, S. H. Chittajallu, R. Aggarwal, A. M. Namboodiri, S. Reddi

Tell us what your model needs to learn

We set up the program around your setting and tasks.

Write to sales@dreamvu.ai.