G cubed VLA: Geometric Inductive Bias for Vision-Language-Action Models

arXiv:2606.24472 · cs.RO, cs.AI · Submitted 2026-06-23 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: Today's paper: "G cubed VLA".

Dev: The gist The G3VLA framework introduces a camera-aware geometric module that injects calibrated structure into the visual-token stream of pretrained Vision-Language-Action (VLA) policies without altering its action space or imitation…

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So we’re talking about G cubed VLA, which is this framework that injects calibrated camera geometry directly into the visual tokens of a pretrained Vision-Language-Action policy without changing how it actually moves or what its goal is.

Dev: That’s right, Rosa. The core idea here is to give these models explicit spatial information—things like camera positions and directions—that they can use immediately, which is usually something they have to learn implicitly from just looking at the actions.

Taro: So you're saying we're bypassing that slow process of learning geometry solely through action supervision? That sounds like it could really speed things up for spatial tasks.

Rosa: Exactly. They do this by adding three geometric modules: intrinsic-conditioned ray embeddings, projective positional encoding, and bidirectional cross-view fusion. This means they are modifying the visual token stream before the policy even sees it >

Dev: Intrinsic-conditioned ray embeddings are essentially tagging each token with its back-projected viewing direction from the camera intrinsics, which is a big step because it resolves that ambiguity in 2D image coordinates > <ref:2606.24472#pg1>

Taro: And then you have this bidirectional cross-view fusion using projective positional encoding to encode those camera relationships across different views, which lets the model understand how views are geometrically coupled >

Rosa: Right. The paper shows they start by learning a calibration-conditioned transformation to add that structure into the pretrained vision encoder's output tokens >

Dev: And they do this in a two-stage training curriculum, which is interesting because it bootstraps the geometry modules first using only sparse, task-level gradients from the action loss before fine-tuning everything together >

Taro: I wonder how that two-stage process handles when things go wrong during the initial geometric learning phase, especially if we’re dealing with real world data that isn't perfectly calibrated >

Rosa: The distillation objective they use for this is dense, targeting both ray-coordinate predictions and depth predictions from a teacher model, which helps them get that geometric supervision without needing perfect ground truth point maps all the time >

Dev: That means when they are training Stage one they use these teacher targets—either real ground truth points or confidence-gated predictions from a π3X teacher—to guide the geometry learning > <ref:2606.24472#pg1>

Taro: So if we look at the results, it seems like this geometric injection really helps on spatially and object-sensitive tasks when tested on systems like π0 > <ref:2606.24472#pg2>

Rosa: It does. On that system, G cubed VLA shows consistent gains across several suites, with the biggest improvements happening where spatial reasoning and understanding objects are key >

Dev: But the validation isn't perfect either; they tested it on GR00T one point five and found mixed results, which suggests the benefit of this geometric injection really depends on how directly those geometry-aware tokens access the action generation pathway >

Taro: That makes sense. If the tokens have to go through an extra bottleneck before they influence the final action decision, the gain might be less pronounced >

Rosa: It seems that's where it lies, and ablation studies confirm that removing components like ray embeddings or PRoPE does reduce success rates, showing they are complementary parts of this geometric structure >

Dev: And looking at the supervision source ablation, they found that using ground truth point maps gives them the strongest signal for learning geometry, but even the confidence-gated π3X predictions provide a significant boost over baseline models > <ref:2606.24472#pg1>

Taro: It’s good to see that confidence gating is used; it means they aren't just blindly following imperfect teacher targets when things get noisy, which is important for robustness >

Rosa: And the final part they show is that at inference, the system only needs RGB images and camera calibration—it doesn't need to query those auxiliary geometry heads or teacher models again >

Dev: So what this means for deployment is that we don't need a whole new pipeline just to get spatial precision; we can use this as a lightweight inductive bias for spatial accuracy in generalist VLA policies >

Taro: It gives the model an inherent understanding of the physical world structure from the start, which should make it more reliable when it encounters novel layouts or object orientations >

Rosa: Overall, G cubed VLA is presenting calibrated camera geometry as a very useful inductive bias for spatial precision in generalist VLA models >

Dev: So we’re wrapping up on this paper. It shows that injecting geometric structure directly into the visual tokens via ray embeddings and cross-view fusion provides consistent gains when geometry-aware tokens reach the action pathway directly >

Taro: It seems like a solid way to improve spatial reasoning without having to overhaul the entire policy architecture or rely on complex external depth sensors during operation >

Rosa: We'll save that for next time.

The paper's summary: Rosa: So, G cubed VLA is this framework that takes all those camera specs—the intrinsics and extrinsics—and injects them directly into the visual tokens of a vision-language-action model without changing how it moves or what its goal is.

Dev: Right. It’s about adding calibrated structure to the visual stream, but keeping the original action space and imitation objective exactly as they were >

Rosa: The core mechanism involves three things: intrinsic-conditioned ray embeddings, projective positional encoding, and bidirectional cross-view fusion >

Dev: Those ray embeddings tag each token with its back-projected viewing direction from those intrinsics. That’s the first layer of calibration injection >

Rosa: And then you have the PRoPE part which uses camera-calibrated biases to encode the projective relationships between different views, like how two cameras are geometrically coupled >

Dev: So it’s using that known geometry to augment the rotary positional embeddings, instead of just relying on what looks similar in appearance >

Rosa: The training is done in two stages. First, you bootstrap those geometry modules using a dense auxiliary geometry distillation objective from a teacher model >

Dev: Then, in stage two, the full policy fine-tunes everything under the action loss being dominant, with the distillation acting as a weaker regularizer >

Rosa: The key takeaway is that this geometric transfer is most effective when those geometry-aware tokens have direct access to the action generation pathway >

Dev: That’s what makes it interesting. It suggests that spatial precision gains come from making the model inherently aware of physical structure, rather than having to learn it slowly through trial and error on actions alone >

Rosa: And when we look at the results on π0, G cubed VLA shows consistent improvements across several suites and the biggest gains are actually seen on tasks that are very sensitive to spatial reasoning or handling objects > <ref:2606.24472#pg2>

Dev: But you can't ignore the caveats. The validation on GR00T one point five showed mixed results, which points to how much that benefit depends on whether those geometry-aware tokens get a direct path to the action generation pipeline >

Rosa: And ablation studies confirm that dropping components like ray embeddings or PRoPE definitely reduces success rates, proving they are all working together >

Dev: It also showed that using ground truth point maps for supervision gives the strongest signal, but even the teacher predictions help recover a lot of the gain >

Rosa: The limitation is also there; it’s sensitive to calibration drift and relying on a visual geometry teacher whose targets can be imperfect under occlusion or specularities >

Dev: And finally, for deployment, they say at inference, you only need RGB images and camera calibration, meaning no extra heads or teacher models are needed during the actual task execution >

Rosa: So the big picture here is that calibrated camera geometry acts as a lightweight inductive bias for spatial precision in generalist models >

Dev: It’s a way to give the AI an inherent understanding of physical structure right at the start, which should make it much more reliable when it encounters novel layouts or object orientations >

Rosa: But we still need to figure out how robust this is when the robot hits something unexpected outside of the lab environment, that's where our trials come in next >

The paper's improvements: Tom: So, G cubed VLA is this framework that takes all those camera specs—the intrinsics and extrinsics—and injects them directly into the visual tokens of a vision-language-action model without changing how it moves or what its goal is.

Rosa: Right. It’s about adding calibrated structure to the visual stream, but keeping the original action space and imitation objective exactly as they were >

Dev: The core mechanism involves three things: intrinsic-conditioned ray embeddings, projective positional encoding, and bidirectional cross-view fusion >

Rosa: Those ray embeddings tag each token with its back-projected viewing direction from those intrinsics. That’s the first layer of calibration injection >

Dev: And then you have the PRoPE part which uses camera-calibrated biases to encode the projective relationships between different views, like how two cameras are geometrically coupled >

Rosa: The training is done in two stages. First, you bootstrap those geometry modules using a dense auxiliary geometry distillation objective from a teacher model >

Dev: Then, in stage two, the full policy fine-tunes everything under the action loss being dominant, with the distillation acting as a weaker regularizer >

Rosa: The key takeaway is that this geometric transfer is most effective when those geometry-aware tokens have direct access to the action generation pathway >

Dev: That’s what makes it interesting. It suggests that spatial precision gains come from making the model inherently aware of physical structure, rather than having to learn it slowly through trial and error on actions alone >

Rosa: And when we look at the results on π0, G cubed VLA shows consistent improvements across several suites and the biggest gains are actually seen on tasks that are very sensitive to spatial reasoning or handling objects > <ref:2606.24472#pg2>

Dev: But you can't ignore the caveats. The validation on GR00T one point five showed mixed results, which points to how much that benefit depends on whether those geometry-aware tokens get a direct path to the action generation pipeline >

Rosa: And ablation studies confirm that dropping components like ray embeddings or PRoPE definitely reduces success rates, proving they are all working together >

Dev: It also showed that using ground truth point maps for supervision gives the strongest signal, but even the teacher predictions help recover a lot of the gain >

Rosa: The limitation is also there; it’s sensitive to calibration drift and relying on a visual geometry teacher whose targets can be imperfect under occlusion or specularities >

Dev: And finally, for deployment, they say at inference, you only need RGB images and camera calibration, meaning no extra heads or teacher models are needed during the actual task execution >

Rosa: So the big picture here is that calibrated camera geometry acts as a lightweight inductive bias for spatial precision in generalist models >

Dev: It’s a way to give the AI an inherent understanding of physical structure right at the start, which should make it much more reliable when it encounters novel layouts or object orientations >

Rosa: But we still need to figure out how robust this is when the robot hits something unexpected outside of the lab environment, that's where our trials come in next >

Conclusion: Rosa: So we’ve looked at G cubed VLA, which is this framework that injects calibrated camera geometry directly into the visual tokens of a vision-language-action model without changing how it moves or what its goal is >

Dev: Right. It’s about adding calibrated structure to the visual stream, but keeping the original action space and imitation objective exactly as they were >

Rosa: The core mechanism involves three things: intrinsic-conditioned ray embeddings, projective positional encoding, and bidirectional cross-view fusion >

Dev: Those ray embeddings tag each token with its back-projected viewing direction from those intrinsics. That’s the first layer of calibration injection >

Rosa: And then you have the PRoPE part which uses camera-calibrated biases to encode the projective relationships between different views, like how two cameras are geometrically coupled >

Dev: So it’s using that known geometry to augment the rotary positional embeddings, instead of just relying on what looks similar in appearance >

Rosa: The training is done in two stages. First, you bootstrap those geometry modules using a dense auxiliary geometry distillation objective from a teacher model >

Dev: Then, in stage two, the full policy fine-tunes everything under the action loss being dominant, with the distillation acting as a weaker regularizer >

Rosa: The key takeaway is that this geometric transfer is most effective when those geometry-aware tokens have direct access to the action generation pathway >

Dev: That’s what makes it interesting. It suggests that spatial precision gains come from making the model inherently aware of physical structure, rather than having to learn it slowly through trial and error on actions alone >

Rosa: And when we look at the results on π0, G cubed VLA shows consistent improvements across several suites and the biggest gains are actually seen on tasks that are very sensitive to spatial reasoning or handling objects > <ref:2606.24472#pg2>

Dev: But you can't ignore the caveats. The validation on GR00T one point five showed mixed results, which points to how much that benefit depends on whether those geometry-aware tokens get a direct path to the action generation pipeline >

Rosa: And ablation studies confirm that dropping components like ray embeddings or PRoPE definitely reduces success rates, proving they are all working together >

Dev: It also showed that using ground truth point maps for supervision gives the strongest signal, but even the teacher predictions help recover a lot of the gain >

Rosa: The limitation is also there; it’s sensitive to calibration drift and relying on a visual geometry teacher whose targets can be imperfect under occlusion or specularities >

Dev: And finally, for deployment, they say at inference, you only need RGB images and camera calibration, meaning no extra heads or teacher models are needed during the actual task execution >

Rosa: So the big picture here is that calibrated camera geometry acts as a lightweight inductive bias for spatial precision in generalist models >

Dev: It’s a way to give the AI an inherent understanding of physical structure right at the start, which should make it much more reliable when it encounters novel layouts or object orientations >

Taro: I'm just thinking about what happens when the world misbehaves outside of those clean lab settings. Does this injection handle unexpected things like weird lighting or sudden occlusion gracefully?

Rosa: That’s the million-dollar question for me. I’ve been running trials, and while it works well in controlled setups, real-world chaos tests its limits >

Dev: From an engineering standpoint, we have to worry about the loop rate here. If that geometric calculation adds too much latency to the action generation pathway, the whole real-time control system breaks down >

Taro: Exactly. If it doesn't recover gracefully when things go wrong—like if the geometry estimation drifts—then all this spatial awareness is useless when we need robustness.

Rosa: So, that’s why we keep pushing those field trials, trying to see how long this calibration structure holds up before the world throws a curveball at it >

Dev: We need to see if the safety case holds when the geometry input becomes noisy or corrupted by unpredictable visual noise >

Taro: It’s about testing that failure mode, seeing how well the system recovers when those calibrated tokens start feeding it junk data instead of structure.

Rosa: Okay, so we’re leaving G cubed VLA for now. Next up, we’ll look at how other systems are tackling those real-time safety and spatial annotation challenges in the next set of papers.

Yue Peng, Yongzhe Zhao, Artur Habuda, Khuyen Pham, Yanheng Zhu

New York University Shanghai · Technical University of Denmark · MBZUAI - Mohamed bin Zayed University of Artificial Intelligence - Mohamed bin Zayed University of Artificial Intelligence

cs.RO, cs.AI

Submitted: 2026-06-23

Updated: 2026-10-14

Project page: https://sites.google.com/view/g3vla

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 91/100

The gist: The gist The G3VLA framework introduces a camera-aware geometric module that injects calibrated structure into the visual-token stream of pretrained Vision-Language-Action (VLA) policies without

Key concepts

Vision-Language-Action (VLA) Models
These are AI models that combine vision (seeing images), language (understanding text), and action (performing robot movements). They learn to manipulate objects based on what they see and what they read, making them generalist robots.
Camera-Aware Geometric Module
This is the core innovation. It's a new component added to VLA models that uses known camera parameters (like focal length and position) to inject precise 3D spatial information into the visual data stream. This helps the model understand where objects are in real space, not just on a flat 2D image.
Intrinsic-Conditioned Ray Embeddings
These embeddings tag each visual token with its specific viewing direction relative to the camera's internal parameters. By knowing this direction, the model gains a geometric understanding of how it is looking at an object, which is crucial for accurate spatial reasoning in 3D environments.
Bidirectional Cross-View Fusion
This technique allows information from multiple cameras to be exchanged and combined. It uses projective positional encoding (PRoPE) as an attention bias to merge the geometric context from different camera perspectives, creating a consistent, geometrically sound representation across all views.

Terminology

Summary

The gist The G3VLA framework introduces a camera-aware geometric module that injects calibrated structure into the visual-token stream of pretrained Vision-Language-Action (VLA) policies without altering its action space or imitation objective, yielding consistent gains across several VLA suites and demonstrating that geometric transfer is most effective when geometry-aware tokens have direct access to the action generation pathway. <ref:2606.24472#pg14>

How it works

G3VLA injects calibrated structure into the visual-token stream of pretrained VLA policies via three geometric modules: intrinsic-conditioned ray embeddings, projective positional encoding (PRoPE), and bidirectional cross-view fusion <ref:2606.24472#pg14>. The process begins by learning a calibration-conditioned transformation to add structure to the pretrained vision encoder's output tokens <ref:2606.24472#pg14>. This transformation comprises intrinsic-conditioned ray embeddings that tag each token with its back-projected viewing direction from K−1, and bidirectional cross-view fusion which exchanges geometric context across camera streams using PRoPE as a calibration-derived attention bias <ref:2606.24472#pg14>.

Geometry Distillation and Two-Stage Training

The geometric module is initialized from scratch and receives only a sparse, task-level gradient from the action loss <ref:2606.24472#pg14>. To bootstrap the geometric pathway, G3VLA employs a two-stage training curriculum <ref:2606.24472#pg14>.

  1. Stage 1 focuses on Geometric Module Pre-training, where only the ray embeddings, cross-view fusion layers, and auxiliary point head are updated while keeping the pretrained backbone frozen and the distillation loss dominant <ref:2606.24472#pg14>.

  2. Stage 2 involves Full Policy Fine-tuning, starting from the Stage 1 checkpoint where all parameters are unfrozen and the action loss becomes dominant, with distillation serving as a weaker regularizer <ref:2606.24472#pg14>.

Geometry Distillation and Supervision

A dense auxiliary geometry distillation objective is introduced to provide geometric supervision, which targets both ray-coordinate predictions qˆv u and log-z depth predictions ˆd v u <ref:2606.24472#pg14>. This supervision targets come from two sources:

  1. Ground-truth point maps when depth is available, where the validity mask mv u = 1 everywhere <ref:2606.24472#pg14>.

  2. Confidence-gated π3X teacher predictions when only camera intrinsics and extrinsics are provided, using a hard gate where mv u = 1[σ(c v u) > τ] with τ = 0.1 <ref:2606.24472#pg14>.

The distillation loss is unified across both modes as Ldistill = X v,u mv u (1/2) qˆv u − q v u 2 + (ˆd v u − d v u) 2 X v,u +ϵ <ref:2606.24472#pg14>.

Validation and Results

G3VLA was instantiated on three architecturally distinct VLA systems: π0, π0.5, and GR00T 1.5 <ref:2606.24472#pg14>. On π0, G3VLA yields consistent improvements across LIBERO suites and RoboCasa24, with the largest gains on spatially and object-sensitive tasks <ref:2606.24472#pg14>. Validation on GR00T 1.5 showed mixed results, suggesting that the benefit of geometric injection depends on how directly geometry-aware tokens participate in action generation <ref:2606.24472#pg14>. Real-world experiments on a bimanual UR5 workbench show that GT-based variants exhibit robust recovery behavior during the pouring nut task, whereas vanilla models do not <ref:2606.24472#pg14>.

Ablation and Limitations

Ablation studies confirm that removing ray embeddings reduces average success from 87.0% to 85.0%, and removing PRoPE reduces it to 85.9%, confirming the complementary nature of these geometric components <ref:2606.24472#pg14>. The supervision source ablation shows that G3VLA (GT) reaches 88.1% while G3VLA (π3X) reaches 87.0%, indicating that GT supervision provides the strongest signal <ref:2606.24472#pg14>. Limitations include sensitivity to calibration drift and dependence on a visual geometry teacher whose targets stay imperfect under occlusion or specularities <ref:2606.24472#pg14>. Furthermore, gains are attenuated on the two-tower GR00T 1.5 architecture because geometric tokens must cross an extra attention bottleneck before reaching action generation <ref:2606.24472#pg14>.

Conclusion

Overall, calibrated camera geometry is a lightweight and useful inductive bias for spatial precision in generalist VLA policies <ref:2606.24472#pg14>. The results suggest that geometric transfer is most effective when geometry-aware tokens reach the action pathway directly <ref:2606.24472#pg14>.

--- Page 1 ---

G3VLA: Geometric inductive bias for Vision-Language-Action Models <ref:2606.24472#pg14>

Yue Peng Yongzhe Zhao Artur Habuda Khuyen Pham Yanheng Zhu Tran Nguyen Le Fares Abu-Dakka Li Guo Project Leads <ref:2606.24472#pg14>

Figure 1: G3VLA overview. (A) Geometric inductive bias is injected into VLA visual tokens via intrinsic-conditioned ray embeddings (K−1) and bidirectional cross-view fusion with PRoPE, leaving the pretrained backbone and action objective unchanged. (B) Stage 1 distills dense point maps from π3X to pretrain the geometry modules; Stage 2 fine-tunes the full policy under action and distillation losses jointly <ref:2606.24472#pg14>

Abstract: Vision-language-action (VLA) models have made rapid progress in generalist robot manipulation by harnessing semantic knowledge from pretrained vision-language backbones, but their visual tokens remain grounded in 2D image coordinates rather than the calibrated geometry of the robot’s cameras — a mismatch especially pronounced in multi-camera setups, where views are coupled by known intrinsics and extrinsics yet processed as independent images <ref:2606.24472#pg14>. We propose G3VLA, a camera-aware geometric module that injects calibrated structure into the visual-token stream of a pretrained VLA without altering its action space or imitation objective, combining intrinsic-conditioned ray embeddings, projective positional encoding (PRoPE), and bidirectional cross-view fusion <ref:2606.24472#pg14>. Geometric supervision is provided either from ground-truth point maps when available, or from confidence-gated π3X teacher predictions, requiring no depth sensors or manual annotations <ref:2606.24472#pg14>. Instantiated on π0, G3VLA yields consistent gains across the LIBERO suites, RoboCasa24, RoboTwin2.0, and real-robot settings, with the largest improvements on spatially and object-sensitive tasks <ref:2606.24472#pg14>. We further validate on π0.5 and GR00T 1.5, with results suggesting that geometric transfer is most effective when geometry-aware tokens have direct access to the action generation pathway <ref:2606.24472#pg14>.

Keywords: Vision-Language-Action, Geometric Inductive Bias, Camera Geometry <ref:2606.24472#pg14>

arXiv:2606.24472v1 [cs.RO] 23 Jun 2026

--- Page 1 ---

G3VLA: Geometric inductive bias for Vision-Language-Action Models

Yue Peng Yongzhe Zhao Artur Habuda Khuyen Pham Yanheng Zhu Tran Nguyen Le Fares Abu-Dakka Li Guo Project Leads

Figure 1: G3VLA overview. (A) Geometric inductive bias is injected into VLA visual tokens via intrinsic-conditioned ray embeddings (K−1) and bidirectional cross-view fusion with PRoPE, leaving the pretrained backbone and action objective unchanged.

Improvements for AI systems

  1. textbfIntrinsic-Conditioned Ray Embeddings for View Direction Awareness: Injecting Calibration into Tokens via Eq. 4 and Eq. 3 (Ray Map Definition). These embeddings tag each ViT patch token with its back-projected viewing direction from K−1, resolving the ambiguity of 2D positional embeddings by incorporating the calibrated intrinsic matrix directly into the visual-token stream before cross-view fusion. This allows downstream attention to operate on tokens that are intrinsically aware of their viewing direction, which is critical for spatial reasoning.

  2. textbfProjective Positional Encoding (PRoPE) for Cross-View Geometric Reasoning: PRoPE augments rotary positional embeddings with a camera-calibrated bias encoding cross-view projective relationships, effectively allowing the model to encode camera-model-based projective relations rather than relying on appearance similarity alone. This mechanism enables the model to leverage known intrinsics and extrinsics to understand geometric coupling between different views, which is essential for multi-camera setups.

  3. textbfTwo-Stage Training Curriculum for Geometry Learning: The paper introduces a curriculum where Stage 1 focuses on bootstrapping the geometry modules with dense auxiliary geometry distillation objective from a teacher, followed by Stage 2 where the action loss dominates. This approach ensures that geometry-aware tokens have direct access to the action generation pathway, leading to gains on spatially and object-sensitive tasks.

  4. textbfConfidence-Gated Teacher Distillation for Geometry Supervision: The method utilizes a distillation objective where supervision targets are selected via a confidence gate: "mv u = 1[σ(c v u) > τ]," allowing the system to learn from teacher predictions only when they exceed a reliability threshold, which prevents the model from being biased by imperfect geometry data. This is further supported by the statement that GT supervision provides the strongest signal, but π3X distillation recovers most of the gain over baseline.

  5. textbfDirect Geometric Inference at Inference: The system is designed so that at inference, the policy requires only RGB images, proprioceptive state, a language instruction, and camera calibration—neither π3X nor the auxiliary head is queried. This ensures that calibrated geometry is used as a lightweight and useful inductive bias for spatial precision without requiring complex auxiliary heads or teacher models during deployment.

Abstract

Vision-language-action (VLA) models have rapidly advanced generalist robot manipulation by harnessing the semantic knowledge of pretrained vision-language backbones. However, their visual representations are typically grounded in 2D image coordinates rather than calibrated camera geometry. This limitation is especially pronounced in multi-camera systems, where views are coupled by known intrinsics and extrinsics but processed as independent images. We propose G cubed VLA, a camera-aware geometric module that injects calibrated structure into the visual-token stream of a pretrained VLA without altering its action space or original action-learning objective. G cubed VLA combines intrinsic-conditioned ray embeddings, projective positional encoding (PRoPE), and bidirectional cross-view fusion to establish geometrically consistent representations across camera views. Geometric supervision is provided by ground-truth point maps or confidence-gated π cubed X teacher predictions, enabling training without depth sensors or manual geometric annotations. On π 0, G cubed VLA consistently improves performance on LIBERO and RoboCasa24 under both supervision regimes, while ground-truth supervision improves five-task RoboTwin2.0 averages with three cameras. Experiments on camera-shifted LIBERO-Goal and real robots demonstrate improved robustness to unseen camera viewpoints. Controlled calibration studies across π 0 and π 0.5 show that these gains depend on correct camera geometry while remaining stable under mild calibration perturbations. These results demonstrate that grounding VLA visual tokens in calibrated camera geometry provides a simple and general mechanism for improving geometric consistency, multi-view reasoning, and viewpoint robustness in pretrained robot policies.

Sources

Related papers