THAILAND FILM OFFICE LICENSED FILM COORDINATOR
Robotics Training Data Collection & Multi-View Video Acquisition Thailand | BVP
BVP Robotics Data Aquisition Division provides synchronized multi-view video data acquisition for robotics, Physical AI, VLA and multimodal model training in real-world workplaces across Thailand.
Synchronized multi-view video capture, structured metadata and production infrastructure for robotics, Physical AI and multimodal model training.
Bangkok Video Productions provides large-scale real-world data acquisition for robotics and AI companies requiring human task demonstrations inside genuine working environments.
We work inside operating factories, warehouses, logistics centers, retail environments, hospitality businesses and other commercial facilities to record employees carrying out normal work activities.
Our role is not simply to film video.
We design and operate a repeatable technical data-acquisition system built around synchronized wearable cameras, external cameras, structured metadata, quality control, secure storage and client-defined delivery standards.
Robotics systems learn from real-world interaction.
BVP captures workers performing genuine tasks in their normal working environment, including:
picking, packing and sorting l assembly and manufacturing l tool use
material handling l warehouse operations l food preparation
retail activities l hospitality workflows l repetitive hand tasks
object manipulation l equipment operation l multi-step workplace processes
Participants are genuine employees performing their normal work rather than actors recreating tasks for camera.
This allows our clients to acquire natural human-action data for robotics, embodied AI, Vision-Language-Action systems and multimodal model development.
Following our technical evaluation, BVP's baseline multi-view acquisition system is built around six matched GoPro HERO13 Black cameras.
A typical configuration includes:
Head-mounted camera
Captures the worker's forward visual perspective and primary field of view.
Left wrist / forearm camera
Captures left-hand movement, object interaction and manipulation.
Right wrist / forearm camera
Captures right-hand movement, object interaction and manipulation.
Three external cameras
Provide wide and medium third-person perspectives of the worker, task area, body movement and surrounding environment.
The cameras record their full-resolution master files locally to high-capacity microSD media. Wireless connectivity is used for camera configuration, monitoring, control and preview rather than relying on Wi-Fi as the primary recording method.
Using the same camera model throughout the system simplifies camera settings, batteries, recording media, mounts, spares, operator training and quality control.
Ultra Wide Lens Mods can be fitted to wearable cameras where a wider field of view is beneficial.
The GoPro HERO13 Black is already being used specifically for robotics data acquisition.
Trossen Robotics specifies the GoPro HERO13 Black with Ultra Wide Lens Mod for its TRumi robot-learning data-collection system.
The Trossen workflow uses the camera to capture wrist-mounted visual observations, combines the recordings with visual-inertial SLAM and extracts synchronized frames, end-effector pose and gripper data for structured Zarr or MCAP datasets.
This provides a documented commercial robotics precedent for the camera platform selected by BVP.
Synchronization remains one of the most important elements of multi-view robotics data acquisition.
For our GoPro HERO13 Black configuration, the baseline workflow can use GoPro Labs precision time and timecode synchronization, together with a recorded synchronization event visible to all cameras.
Available synchronization methods include:
GoPro Labs precision time/timecode l QR-based camera time synchronization l GPS-assisted timecode synchronization where appropriate l LTC timecode input or jam-sync using compatible hardware l simultaneous visual synchronization event l synchronization verification during ingest and QC
A visible synchronization event can be recorded by every camera at the beginning and end of an acquisition session, providing an additional independent reference during data processing.
The synchronization specification is agreed with the client's robotics engineering team before collection begins.
For projects requiring approximately frame-level synchronization, the HERO13 workflow provides a practical and proven solution.
Where a client's engineering specification requires sub-frame, millisecond-level, hardware-triggered, PTP or genlock synchronization, BVP can deploy a different specialist camera and sensor architecture designed specifically for that requirement.
We therefore select the camera architecture according to the synchronization tolerance required by the client's training pipeline rather than applying one camera system to every project.
BVP's proposed software workflow is based around widely adopted robotics technologies rather than proprietary video-production software. Our baseline stack can include:
ROS 2:
Used where integration with robotics middleware, sensors or client robotics systems is required.
MCAP:
Our preferred container format for timestamped multimodal robotics data.
MCAP can combine camera streams, timestamps, metadata, sensor information and other structured data into a format designed specifically for robotics workflows.
Foxglove:
Used for visual inspection, synchronized playback, troubleshooting and quality control of robotics data.
Foxglove allows technical teams to inspect multiple data streams against a common timeline and identify synchronization, camera, metadata or sensor problems before data is accepted.
LeRobot Dataset:
Where appropriate, accepted sequences can also be structured for compatibility with the Hugging Face LeRobot Dataset format. LeRobot is specifically designed for robot-learning datasets containing multiple camera observations, time-series data, actions, robot states and associated metadata.
Rerun:
Rerun can also be incorporated for local visualization and inspection of multimodal robotics datasets.
Client-Defined Formats:
The client's own engineering specification always takes priority. BVP can deliver according to a proprietary schema, folder structure, naming convention, metadata requirement, cloud-ingestion specification or machine-learning pipeline.
Proven Industry Adoption:
These technologies are already being used by major robotics and Physical AI organizations. Trossen Robotics produces MCAP data through its TRumi robot-learning data-collection pipeline.
NVIDIA Isaac ROS incorporates MCAP for multi-sensor robotics data recording.
Foxglove identifies robotics and Physical AI users including NVIDIA, Dexterity, ANYbotics, Wayve, Shield AI and Waabi. Hugging Face's LeRobot Dataset architecture has been developed specifically for large-scale robot-learning datasets, including synchronized multi-camera observations.
This allows BVP to build its acquisition workflow around technologies already being used within the robotics industry rather than developing a proprietary data format that clients must adapt to.
We recommend retaining two layers of project data.
The original camera recordings are retained without modification.
These provide a complete source record of the acquisition session.
Accepted sequences can then be organized with synchronized timing and associated metadata for use by the client's robotics or machine-learning pipeline.
This approach provides both:
high-quality original source media and structured machine-readable training data.
Video alone is often insufficient for robotics training.
BVP can create structured metadata for each acquisition session including:
session ID l facility l environment type l employee/participant ID l task category l activity description l camera ID l camera position l camera type l recording start/end time l synchronized sequence duration l object interaction l workplace description l participant count l task result l acceptance status l notes and exceptions
Additional client-defined metadata can be incorporated before production begins.
Before beginning a major robotics data-acquisition program, BVP recommends a dedicated three-day technical training and validation program.
The production team is trained in:
GoPro HERO13 Black configuration l camera identification and standardized settings l head and wrist camera fitting l external camera positioning l field-of-view requirements l exposure and image-quality standards l battery and recording-media management l wireless camera control and monitoring l timecode configuration l synchronization procedures l synchronization verification l recording start and stop procedures l camera fault and replacement procedures
The team is trained in:
project dataset structure l ROS 2 workflow where required l MCAP data organization l Foxglove inspection and quality control l Rerun visualization where applicable l LeRobot-compatible dataset structures l session and episode creation l metadata entry l file naming l camera identification l data ingest l checksum verification l synchronization checking l rejection criteria l dataset versioning l NAS and backup procedures l client data-transfer procedures
The complete team then performs a supervised acquisition inside a genuine working environment.
The field test validates:
worker comfort l head-camera position l wrist-camera position l external camera coverage l camera synchronization l task visibility l lighting l normal worker movement l recording reliability l metadata collection l data ingest l software processing l Foxglove / dataset QC l backup procedures l rejected-sequence workflow l client review procedure
Only after the complete camera, software, synchronization and data pipeline has been successfully tested do we move into large-scale production.
BVP can build a dedicated hardware package for each long-term project.
Typical equipment includes:
matched action cameras l head mounts
left and right wrist mounts l external cameras
tripods and clamps l batteries and charging stations
high-capacity recording media l synchronization equipment
ingest workstation l high-speed card readers
dedicated NAS storage l backup drives
secure networking l remote transfer infrastructure
For long-term assignments, equipment can be dedicated exclusively to the client's program.
Large-scale robotics acquisition can generate hundreds of terabytes of data.
Our workflow can include:
Cameras
↓
On-site data ingest
↓
Checksum / file integrity verification
↓
BVP project NAS
↓
Automated client storage replication
↓
Quality control and sequence validation
↓
Accepted data delivery
The BVP NAS can function as a local working and backup repository while client-approved data is transferred to the client's own infrastructure.
Client destinations can include:
private servers l NAS systems l AWS S3 l Microsoft Azure l Google Cloud Storage l client-controlled object storage l secure data-transfer servers
Transfer architecture is agreed with the client's technical team before production.
Every robotics acquisition program should have a defined acceptance specification.
BVP can conduct QC checks covering:
missing cameras l corrupted media l dropped recordings
synchronization l frame rate l resolution
camera obstruction l camera movement l incorrect wrist/head position
exposure l usable field of view l incomplete tasks
metadata completeness l file naming l session duration
checksum verification
Only sequences meeting the agreed specification are submitted as accepted usable data.
For robotics work, we do not simply measure production by the number of hours the cameras are switched on.
Data can be organized into individual:
sessions l tasks l episodes l sequences
A synchronized sequence may include simultaneously recorded:
head perspective
left wrist
right wrist
external camera 1
external camera 2
external camera 3
Together these streams represent one synchronized multi-view sequence.
This structure makes the dataset easier for robotics engineers to inspect, curate and incorporate into model-training pipelines.
We can source and coordinate data acquisition in a wide range of Thailand environments.
Factories, production lines, assembly, electronics, automotive, food processing and industrial workplaces.
Picking, packing, sorting, pallet handling, inventory movement and fulfillment operations.
Shelf stocking, checkout, inventory, customer-service tasks and store operations.
Hotels, restaurants, kitchens, housekeeping and food-service workflows.
Offices, workshops, service businesses and specialized commercial environments.
BVP can manage the operational side of the project as well as the technical acquisition.
This can include:
facility sourcing / facility agreements / employee coordination
participant documentation / releases / scheduling
production management / safety coordination / confidentiality procedures
secure file handling / long-term data-acquisition staffing
For large projects, we can establish a dedicated monthly production team rather than operating with conventional film-industry day crews.
BVP can structure programs from an initial technical pilot through to hundreds or thousands of accepted synchronized data hours.
A typical rollout may include:
Phase 1 — Technical Planning
Client specifications and acceptance standards.
Phase 2 — Training
Classroom and practical team training.
Phase 3 — Pilot Acquisition
Real-world data collection and client engineering review.
Phase 4 — Workflow Validation
Synchronization, metadata, storage and acceptance verification.
Phase 5 — Scaled Acquisition
Dedicated production team collecting data against monthly targets.
Phase 6 — Continuous QC & Delivery
Accepted sequences transferred to client infrastructure throughout the project.
Large-scale robotics data collection requires more than conventional camera operators and production crew.
For long-term programs, BVP's technical structure includes a dedicated Robotics Data Acquisition & Software Lead responsible for the complete technical data pipeline.
This specialist works between the BVP production team and the client's robotics, machine-learning and data-engineering teams.
Primary responsibilities include:
configuring the robotics data-acquisition software stack / establishing the MCAP and metadata architecture
implementing synchronization and timestamp procedures / designing camera and data-ingest workflows
supervising camera and sensor calibration / developing automated data validation and QC procedures
managing Foxglove and Rerun inspection workflows / preparing LeRobot-compatible datasets where required
maintaining dataset naming and version control / supervising checksum, storage and backup procedures
investigating synchronization or data-quality failures / producing technical SOPs
training camera and data-acquisition operators / communicating directly with client robotics engineers
Master's degree in Robotics, Computer Science, Software Engineering, Mechatronics or a related technical discipline, or equivalent professional experience.
The position requires professional experience with:
robotics or autonomous systems / ROS 2 / Python / Linux / MCAP / rosbag / multi-camera or multi-sensor data pipelines / camera and sensor synchronization / timestamp management / camera calibration / structured metadata / robotics dataset QC / large data storage and transfer / Git and software version control
Experience with Foxglove, Rerun, LeRobot, C++, Docker, cloud storage, SLAM, computer vision and machine-learning data pipelines would be highly desirable.
This role provides technical ownership of the acquisition pipeline and ensures that BVP is delivering structured robotics training data — not simply recording multiple video files.
The Founder of BVP has extensive experience in software application development pioneering Thailand's development of the fist wave of Internet based global-interbank banking and securities trading software with an additional 20 years experience in the film industry capturing high qualuity visual data.
This unique combination provides the knowledge of both software with the importance of capturing high quality video footage & associated data. View Profile
Bangkok Video Productions combines more than conventional camera production.
For robotics and Physical AI projects, we bring together:
Real-world production logistics
Access to genuine workplaces and operational facilities.
Professional multi-camera acquisition
Wearable and external synchronized camera systems.
Technical production management
Repeatable procedures designed for large-volume projects.
Data infrastructure
High-capacity storage, structured ingest, backup and secure delivery.
Quality control
Acceptance-based data review before final delivery.
Thailand production capability
Local crews, facility sourcing and nationwide production coordination.
Long-term scalability
Dedicated teams capable of supporting continuous data-acquisition programs.
Tell us what your model needs to learn.
Send us your required tasks, environment types, camera specification, synchronization tolerance, metadata schema and target number of accepted data hours.
BVP can develop a technical acquisition plan, pilot program and scalable production workflow around your requirements.
REQUEST A ROBOTICS DATA ACQUISITION QUOTE
International Telephone: +66 89518 6076 (English)
Local Telephone: 084 945 5442 (Thai & English)
International WhatsApp: +66 89518 6076 (English)
Local WhatsApp: 084 945 5442 (Thai & English)
International Line ID: bangkokvideo (English)
Local Line ID: JennyBalee (Thai & English)
International Telegram: @bangkokvideo (English)
WeChat: +66 89518 6076 (English)
