# Independent Variable: Convergence of Belief Is Not Convergence of Evidence

Independent Variable: Convergence of Belief Is Not Convergence of Evidence

Full article: https://haiguangboy.com/posts/xsquare-robot

## Core Approach

↳ What you're buying is a belief that is being independently verified by multiple parties
①CEO Wang Qian bets on "building the brain, not the body," rejecting the marathon-running robot track; this judgment heavily overlaps with Xu Huazhe, founder of Poke, who independently reached "no robotics-style single-task optimization" around the same time
②WALL-B's WUM architecture (unified network for vision/language/action/physical prediction) aligns with Xu Huazhe's "action layer must be a unified model" judgment, and echoes the current architectural shift among multiple companies abandoning VLA modular splicing
③The real logic behind the 149-yuan cleaning service is "deploy-to-collect"—using aunties and engineers as scaffolding to get models that are far from ready into real homes now to generate data; this strategy shares the same origin as Sunday's "single-point deployment to start the flywheel" logic
④The judgment of "building the brain, not the body" also matches third-party analysts' industry observation that "bipedal is constrained in the short term by motion control, wheeled scales first"—the independent variable is precisely self-developing wheeled dual-arm robots, not bipedal humanoids

## Key Results

Four parties independently reached similar conclusions, which looks like multiple validation, but it's worth distinguishing: this is more like several people converging on the same bet based on the same limited public information, not four independent empirical results.

## Related Route Comparisons

- Opposite route · [An_Open_Foundation_Model_Towards](https://haiguangboy.com/posts/an_open_foundation_model_towards) `an_open_foundation_model_towards_2026_07`: Architecture bet: unified network outperforms modular splicing—WUM opposes VLA's semantic layer-by-layer transmission contradicts joint co-training of heterogeneous human-robot data is a structurally suboptimal solution
- Same route · [D4RT_Efficiently_Reconstructing_Dynamic_Scenes_One](https://haiguangboy.com/posts/d4rt_efficiently_reconstructing_dynamic_scenes_one) `d4rt_efficiently_reconstructing_dynamic_scenes_one_2026_06`: Boundary: no published robot task completion rate/independent contribution share; currently more like a data expedition than commercial deployment validates evaluation based on curated benchmarks, field only qualitative; relies on internal data, quality as a 'data source' not quantified
- Same route · [latepost_xuhuazhe_202603](https://haiguangboy.com/posts/latepost_xuhuazhe_202603) `latepost_xuhuazhe_202603_2026_03`: Route bet: build the brain, not the body—betting on foundation model company identity, rejecting the motion capability track validates route bet: AI-native triple negation—not robotics/not autonomous driving/not prehistoric deep learning
- Same route · [sunday_blog_20260717](https://haiguangboy.com/posts/sunday_blog_20260717) `sunday_blog_20260717_2026_07`: Boundary: no published robot task completion rate/independent contribution share; currently more like a data expedition than commercial deployment validates source boundary: self-reported, self-built evaluation, no third party, single task family
- Same route · [π0.5 shakes in place when failing to grab a spoon, but Orca's physical intuition, learned from watching videos, goes further](https://haiguangboy.com/posts/orca) `orca_2026_07`: The key to a world model is readable states
- Same route · [No need to imagine the future during inference, the robot still reaches 91.8%! Fast-WAM debunks WAM's core assumption](https://haiguangboy.com/posts/fast-wam) `fast-wam_2026_07`: Training video objectives matter more than imagining the future at test time

## Boundaries

· 149 yuan is not accountable commercial revenue; it's data collection cost packaged as a consumer service
· The official side has never disclosed any quantitative metrics such as robot independent completion rate or task success rate
· This is not an isolated case—in the same batch of articles, Poke, Sunday, Genesis, and D4RT, none provided third-party verified numbers; all are self-reported

## Author's Judgment (not paper content, cross-paper comprehensive view)

The gathering of the four giants looks more like the same narrative adopted four times, not a coincidence of four independent due diligence efforts—what could truly turn "convergence of belief" into "convergence of evidence" is a number no one has yet published: whether the intervention rate of scaffolding (aunties/engineers) can decline over time. Until that curve emerges, the 149 yuan buys more of an entry ticket than a moat.