Figure Project Go-Big vs GEN-0

Project Go-Big is a humanoid pretraining data-collection initiative that Figure announced in a company blog post on September 18, 2025. Figure says it is building what it calls the world’s largest and most diverse humanoid pretraining dataset for Helix, accelerated by a partnership with Brookfield. GEN-0 is a family of embodied foundation models announced in a Generalist AI blog post, cited as November 2025, and pretrained on the company’s in-house robotics dataset.

The headline difference is the data. Project Go-Big collects egocentric human video. GEN-0 is pretrained on robot manipulation data from Generalist’s own operations. Neither is a Firsthand product, and Firsthand has no relationship with Figure, Brookfield or Generalist. For the single-source explainers, see /guides/what-is-figure-project-go-big and /guides/what-is-gen-0.

Figure states that Project Go-Big is accelerated by a partnership with Brookfield, which it describes as owning over 100,000 residential units worldwide, alongside 500 million square feet of commercial office space and 160 million square feet of logistics space. It says data collection has already begun in Brookfield environments and will continue to scale.

The post reports one result. Figure trained Helix on 100% egocentric human video, collected passively as people do behaviors in real Brookfield homes, with no robot demonstrations, and Helix now responds to commands such as “Walk to the kitchen table” by generating closed-loop navigation control from pixels. Figure says that, to its knowledge, this is the first time a humanoid has learned end-to-end from images and language to low-level SE(2) velocity commands using only human video.

The post does not state an hour count, an episode count, an action-label method, a licence or a benchmark number, so this page does not either.

The GEN-0 post calls it “a new class of embodied foundation models built for multimodal training directly on high-fidelity raw physical interaction”, pretrained on “our in-house robotics dataset, which includes over 270,000 hours of real-world diverse manipulation data, growing at a rate of 10,000 hours a week”. It describes the data as collected across diverse activities in 1,000s of homes, warehouses and workplaces worldwide, using 1,000s of data collection devices and robots.

The post states that more pretraining data and compute consistently and predictably improve downstream post-training performance, with a power-law relationship between pretraining data size and downstream error across 16 task sets. It adds that data quality and diversity matter more than sheer volume. It does not describe its capture rigs beyond custom hardware, does not give an annotation schema, and does not state a licence.

Each supports its own side, and neither tests the other. Project Go-Big’s reported result is a navigation capability learned with no robot data at all. GEN-0’s reported results come from pretraining on robot manipulation data and then post-training, tested on 6DoF, 7DoF and 16+DoF semi-humanoid robots. Neither post compares human video with robot data on the same task.

The trade-off is the one /compare/human-video-vs-robot-data-for-pretraining describes. Robot data is recorded on the target embodiment but needs hardware and operators, and the GEN-0 post describes custom hardware and infrastructure built for its scale. Human video avoids that cost, and Figure argues a humanoid’s perspective and kinematics mirror a human’s, so it can transfer. Read the two as independent evidence for different routes, not as a head-to-head result.

No. Figure’s result is zero-shot navigation from human video alone. GEN-0’s results concern post-training performance on manipulation task sets, including real-robot runs where models post-trained on 5.6 hours of task-specific data improved with more pretraining. Different tasks, different embodiments and different evaluation setups mean the navigation result and the 99% peak figure answer different questions.

GEN-0 states over 270,000 hours of manipulation data. Project Go-Big states property access, not hours. Treat them as different units. Property scale describes how many real spaces a partnership opens up, while an hour count describes how much recorded data exists. A buyer should not convert one into the other, and neither is a target for a custom collection.

Both posts point the same way on diversity. GEN-0 states quality and diversity matter more than sheer volume. Project Go-Big frames diversity of real-world spaces as its goal. A pretraining plan should start from your robot’s coverage gaps. See /guides/why-dataset-diversity-matters-in-robot-learning.

Project Go-Big is Figure’s own initiative for Figure’s own model, built on a property partnership specific to Figure and Brookfield. GEN-0’s dataset is in-house, and the post describes no way for outside teams to obtain it. A team that needs data scoped to its own robot, tasks and environments has to source it separately.

A custom collection starts with a written skill spec covering the scenes, tasks, objects and people a corpus is missing, proven with a fast, on-demand Spec Pilot before volume scales. Capture is head-worn, bimanual and whole-body for humanoid pretraining, delivered under a perpetual, buyer-owned commercial licence with a documented consent chain, in RLDS, LeRobot, WebDataset, HDF5, zarr and Rerun .rrd. See /custom-collection/humanoid-pretraining-data and /custom-collection/pretraining-data-collection.

Ask what the data is (human video or robot data), how action labels were produced, whether the viewpoint matches your robot, whether sensor streams are synchronized, how coverage was planned against a spec, and whether the licence permits commercial training with documented consent. Neither post answers those for outside buyers. See /guides/robot-pretraining-data-diversity-plan.