REVIEW 2 major objections 1 minor 53 cited by
Reviewed by Pith at T0; open to challenge.
T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →
T0 review · grok-4.3
Gemini Robotics 1.5 adds motion transfer and interleaved language reasoning to let multi-embodiment robots handle complex physical tasks.
2026-05-16 07:33 UTC pith:RW4CX7AT
load-bearing objection Gemini Robotics 1.5 adds a motion transfer mechanism for multi-embodiment data and interleaves natural language reasoning in the VLA loop, but the abstract gives no metrics or ablations to check the generalization claims. the 2 major comments →
Gemini Robotics 1.5: Pushing the Frontier of Generalist Robots with Advanced Embodied Reasoning, Thinking, and Motion Transfer
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
A novel architecture equipped with a Motion Transfer mechanism lets the VLA model absorb heterogeneous data from multiple robot embodiments, while interleaving actions with internal natural-language reasoning steps improves decomposition of complex tasks and produces more interpretable behavior; the separate ER model then sets new performance records on the specific reasoning skills required for physical interaction.
What carries the argument
Motion Transfer (MT) mechanism that transfers learned motion patterns across different robot embodiments, combined with multi-level internal reasoning expressed in natural language before each action.
Load-bearing premise
Benchmark gains from motion transfer and interleaved reasoning will carry over to unstructured real-world settings containing objects, lighting, and dynamics absent from training data.
What would settle it
Place the robot in a previously unseen room with novel objects and changed lighting, then measure whether it still completes the same multi-step tasks it succeeded on in controlled benchmarks.
If this is right
- Robots become able to break down and carry out longer sequences of actions without hand-crafted scripts.
- Behavior becomes more transparent because the internal reasoning chain is expressed in readable language.
- A single model can be deployed on robots with different physical forms after training on mixed data.
- Embodied reasoning benchmarks improve on visual grounding, spatial relations, and step-by-step planning.
Where Pith is reading between the lines
- The same motion-transfer approach could shorten the time needed to adapt the model to entirely new hardware platforms.
- Visible language reasoning opens the possibility of real-time human correction during execution.
- If the reasoning layer generalizes, similar interleaving might improve other embodied agents such as autonomous vehicles or manipulators in warehouses.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces Gemini Robotics 1.5, a multi-embodiment Vision-Language-Action (VLA) model with a novel architecture and Motion Transfer (MT) mechanism designed to learn from heterogeneous robot data, along with interleaved multi-level natural language reasoning to enable 'thinking before acting' for complex tasks. It also presents Gemini Robotics-ER 1.5 as achieving state-of-the-art performance on embodied reasoning benchmarks covering visual/spatial understanding, task planning, and progress estimation. The overall goal is advancing generalist physical agents capable of perception, reasoning, and dexterous control.
Significance. If the claimed generalization benefits from MT and the interleaved reasoning hold under rigorous testing, the work would mark a meaningful advance in multi-embodiment VLAs by addressing embodiment-specific data heterogeneity. The emphasis on interpretable internal reasoning is a positive direction for robot transparency. However, the absence of any quantitative metrics, ablation studies, or cross-embodiment transfer results in the provided text leaves the central performance claims unverified and limits assessment of whether MT genuinely enables embodiment-agnostic representations beyond what larger data or model scale would achieve.
major comments (2)
- [Abstract] Abstract: The central claim that the Motion Transfer (MT) mechanism 'enables it to learn from heterogeneous, multi-embodiment robot data and makes the VLA more general' is not supported by any ablation studies isolating MT's contribution, cross-embodiment transfer metrics (e.g., success rates when training on one embodiment and testing on another), or description of the latent alignment procedure. Without these, benchmark gains cannot be confidently attributed to MT rather than data volume or architecture scale.
- [Abstract] Abstract: The assertion that Gemini Robotics-ER 1.5 'establishes a new state-of-the-art for embodied reasoning' and that the overall family 'takes us a step towards an era of physical agents' is presented without any quantitative results, baseline comparisons, or evaluation protocols. This renders the performance and generalization claims unverifiable from the manuscript as presented.
minor comments (1)
- [Abstract] The abstract uses several forward-looking phrases ('pushing the frontier', 'era of physical agents') that could be toned down to focus strictly on the technical contributions and measured results.
Simulated Author's Rebuttal
We thank the referee for their careful reading and constructive comments on the abstract. We address each point below and will revise the manuscript to strengthen the presentation of our claims.
read point-by-point responses
-
Referee: [Abstract] Abstract: The central claim that the Motion Transfer (MT) mechanism 'enables it to learn from heterogeneous, multi-embodiment robot data and makes the VLA more general' is not supported by any ablation studies isolating MT's contribution, cross-embodiment transfer metrics (e.g., success rates when training on one embodiment and testing on another), or description of the latent alignment procedure. Without these, benchmark gains cannot be confidently attributed to MT rather than data volume or architecture scale.
Authors: We agree that isolating the contribution of MT requires explicit ablations and cross-embodiment transfer results. The full manuscript describes the MT architecture and latent alignment procedure in detail and provides qualitative demonstrations of multi-embodiment learning. However, we acknowledge the absence of quantitative ablations and transfer metrics in the current version. We will add a dedicated ablation study section reporting success rates for training on one embodiment and evaluating on others, along with comparisons to scale-only baselines. revision: yes
-
Referee: [Abstract] Abstract: The assertion that Gemini Robotics-ER 1.5 'establishes a new state-of-the-art for embodied reasoning' and that the overall family 'takes us a step towards an era of physical agents' is presented without any quantitative results, baseline comparisons, or evaluation protocols. This renders the performance and generalization claims unverifiable from the manuscript as presented.
Authors: The abstract summarizes results that are quantified in the main body, where Gemini Robotics-ER 1.5 is evaluated on embodied reasoning benchmarks with direct baseline comparisons and described evaluation protocols. We will revise the abstract to include specific quantitative improvements (e.g., accuracy deltas on visual/spatial, planning, and progress estimation tasks) and a brief reference to the evaluation section so that the SOTA claim is verifiable from the abstract alone. revision: yes
Circularity Check
No significant circularity in the derivation chain
full rationale
The paper introduces Gemini Robotics 1.5 as a multi-embodiment VLA model featuring a Motion Transfer mechanism and interleaved natural-language reasoning, plus a separate Embodied Reasoning model. All central claims are supported by descriptions of training procedures and empirical benchmark results rather than mathematical derivations, equations, or self-referential definitions. No steps reduce predictions or uniqueness claims to fitted inputs or prior self-citations by construction; the argument chain remains self-contained through external data and evaluation.
Axiom & Free-Parameter Ledger
read the original abstract
General-purpose robots need a deep understanding of the physical world, advanced reasoning, and general and dexterous control. This report introduces the latest generation of the Gemini Robotics model family: Gemini Robotics 1.5, a multi-embodiment Vision-Language-Action (VLA) model, and Gemini Robotics-ER 1.5, a state-of-the-art Embodied Reasoning (ER) model. We are bringing together three major innovations. First, Gemini Robotics 1.5 features a novel architecture and a Motion Transfer (MT) mechanism, which enables it to learn from heterogeneous, multi-embodiment robot data and makes the VLA more general. Second, Gemini Robotics 1.5 interleaves actions with a multi-level internal reasoning process in natural language. This enables the robot to "think before acting" and notably improves its ability to decompose and execute complex, multi-step tasks, and also makes the robot's behavior more interpretable to the user. Third, Gemini Robotics-ER 1.5 establishes a new state-of-the-art for embodied reasoning, i.e., for reasoning capabilities that are critical for robots, such as visual and spatial understanding, task planning, and progress estimation. Together, this family of models takes us a step towards an era of physical agents-enabling robots to perceive, think and then act so they can solve complex multi-step tasks.
Forward citations
Cited by 53 Pith papers
-
RobotValues: Evaluating Household Robots When Human Values Conflict
RobotValues is a benchmark of 10K value-conflict scenarios that reveals VLMs default to safety and accommodation while failing to follow instructions to prioritize other values 80% of the time.
-
Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics
Open-H-Embodiment is the largest open multi-embodiment medical robotics dataset, used to train GR00T-H, the first open vision-language-action model that achieves end-to-end suturing completion where prior models fail.
-
Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding
Molmo2 delivers state-of-the-art open-weight video VLMs with new grounding datasets and training methods that outperform prior open models and match or exceed some proprietary ones on pointing and tracking tasks.
-
Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robots
Embodied.cpp introduces a portable C++ inference runtime with modular layers for deploying VLA and WAM models on heterogeneous robots, reporting 100% and 91% task success on two models plus memory reduction on a WAM b...
-
FTP-1: A Generalist Foundation Tactile Policy Across Tactile Sensors for Contact-Rich Manipulation
FTP-1 is the first foundation tactile policy pretrained on ~3000 hours of data from 26 sources across 21 sensors that improves performance on seen setups by 17.2% and transfers to unseen sensors with 31% success rate gain.
-
Beyond Attack Success Rate: Examining Trigger Leakage in Vision-Language Agentic Systems
Introduces Neighbor Leakage Rate showing high trigger leakage in VLAS backdoors at 3% poisoning, caused by broad activation regions in fine-tuning that hard-negative samples can narrow.
-
Pause and Think: A Dataset and Benchmark for Video-Grounded Assistive Action Suggestion
Introduces pause-and-think-T dataset and pause-and-think-B benchmark; fine-tunes 4B VLM to 58% accuracy matching 235B model while generalizing out-of-distribution.
-
Spatially Prompted Visual Trajectory Prediction for Egocentric Manipulation
The paper introduces SP-VTP as a new setting for egocentric manipulation, releases the EgoSPT dataset with first-frame spatial annotations, and proposes the SPOT model that outperforms non-prompted baselines on cross-...
-
Offline Policy Evaluation for Manipulation Policies via Discounted Liveness Formulation
A liveness-based Bellman operator enables conservative offline policy evaluation for manipulation tasks by encoding task progression and reducing truncation bias from finite horizons.
-
MemCompiler: Compile, Don't Inject -- State-Conditioned Memory for Embodied Agents
MemCompiler reframes memory use as state-conditioned compilation, delivering relevant guidance via text and latent channels to improve embodied agent performance up to 129% and cut latency 60% versus static injection.
-
MemCompiler: Compile, Don't Inject -- State-Conditioned Memory for Embodied Agents
MemCompiler introduces state-conditioned memory compilation that dynamically selects and compiles relevant memory into text and latent guidance, yielding up to 129% gains over no-memory baselines and 60% lower latency...
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 delivers an open VLA model with new specialized components, datasets, and techniques that outperforms baselines on benchmarks while releasing all weights, code, and data for real-world robot use.
-
KinDER: A Physical Reasoning Benchmark for Robot Learning and Planning
KinDER is a new open-source benchmark that demonstrates substantial gaps in current robot learning and planning methods for handling physical constraints.
-
Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics
A consortium released the largest open medical robotics dataset spanning 50+ institutions and used it to train an open VLA model achieving 25% full suturing completion and a multi-embodiment surgical world model.
-
PhysMem: Scaling Test-Time Memory for Embodied Physical Reasoning
PhysMem enables VLM-based robot planners to learn and verify physical properties through test-time interaction and hypothesis testing, raising success on a brick insertion task from 23% to 76%.
-
RL-VLA$^3$: A Flexible and Asynchronous Reinforcement Learning Framework for VLA Training
RL-VLA³ is an asynchronous RL framework for VLA training that delivers up to 85.2% higher throughput than synchronous baselines while preserving identical sample efficiency and scaling to 256 GPUs.
-
Sequential Planning via Anchored Robotic Keypoints
SPARK reaches 43.7% success on six LIBERO-PRO cells by LLM-generated typed behavior trees plus multi-prompt perception and recovery, more than doubling CaP-Agent0 and VLA baselines.
-
Training Vision-Language-Action Models with Dense Embodied Chain-of-Thought Supervision
ZR-0 is a dual-stream VLA model trained with dense ECoT supervision on 60M frames from 400K trajectories to enable cross-embodiment transfer in simulation and real-world settings.
-
Training Vision-Language-Action Models with Dense Embodied Chain-of-Thought Supervision
ZR-0 trains a dual-stream VLA model on 60 million frames with dense ECoT annotations so that high-level reasoning transfers across single-arm, bimanual, and humanoid embodiments while skipping reasoning at inference.
-
MANGO: Automated Multi-Agent Test Oracle Generation for Vision-Language-Action Models
MANGO uses Generator, Assessor, and Judge agents to create reusable atomic tasks and fine-grained oracles from natural language, evaluated on LIBERO_10 and RoboCasa benchmarks for comparable failure detection with bet...
-
Vesta: A Generalist Embodied Reasoning Model
Vesta is a unified embodied generalist model that outperforms specialist baselines by over 20% on average and improves real-world robotic task success by over 35%.
-
Finetuning Vision-Language-Action Models Requires Fewer Layers Than You Think
VLA models exhibit layer-wise redundancy allowing up to 50% depth compression via training-free CKA-based removal, yielding faster fine-tuning and inference with no performance loss on robot tasks.
-
Playful Agentic Robot Learning
RATs agents generate and solve their own exploratory tasks during play, distill successful code into a skill library, and reuse it to improve held-out task performance by 20.6 and 17.0 points on two benchmarks.
-
Geometric Action Model for Robot Policy Learning
GAM splits a geometric foundation model to enable language-conditioned future geometry prediction and action decoding for robot policies, claiming superior performance on manipulation benchmarks.
-
DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?
DIRECT is a multimodal-context router that allocates test-time compute across chain-of-thought depth, model size, and memory history for VLM embodied planners, improving the success-cost Pareto frontier and matching s...
-
What Matters in Orchestrating Robot Policies: A Systematic Study of Hierarchical VLA Agents
A systematic study of hierarchical VLA agents identifies design principles that improve robot manipulation performance over flat and naive hierarchical baselines in simulation and real-world experiments.
-
Revisiting Embodied Chain-of-Thought for Generalizable Robot Manipulation
ERVLA trains on a 978k-trajectory embodied CoT corpus using reasoning as supervision with dropout, then predicts actions without CoT at test time, reaching 86.9% on LIBERO-Plus and 53.2% on VLABench.
-
Pause and Think: A Dataset and Benchmark for Video-Grounded Assistive Action Suggestion
Introduces pause-and-think-T dataset and pause-and-think-B benchmark for video-grounded assistive action suggestion, enabling a 4B VLM to match larger models on reasoning tasks and generalize to EgoThink and TempCompass.
-
Safe and Steerable Geometric Motion Policies for Robotic Dexterous Manipulation
SafePBDS uses pullback control barrier functions and a task manifold action interface to generate certifiably safe, steerable motions on high-DOF robots from objectives defined on arbitrary geometric spaces.
-
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
VLMs achieve 53-97% on volumetric rearrangement planning but only 6-45% on occlusion and under 7% on reflections in a new 3,034-sample benchmark, with white-box analysis localizing the failure to visual-token merger i...
-
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
VLMs achieve 53-97% on rearrangement planning but only 6-45% on occlusion and under 7% on reflections, with failures localized to visual token compression after the vision encoder.
-
D-VLA: A High-Concurrency Distributed Asynchronous Reinforcement Learning Framework for Vision-Language-Action Models
D-VLA uses plane decoupling and a swimlane pipeline to deliver higher throughput and linear speedup than prior RL frameworks when training billion- and trillion-parameter VLA models on benchmarks like LIBERO.
-
D-VLA: A High-Concurrency Distributed Asynchronous Reinforcement Learning Framework for Vision-Language-Action Models
D-VLA introduces plane decoupling and a swimlane asynchronous pipeline to achieve high-concurrency RL training and linear scalability for billion- to trillion-parameter vision-language-action models.
-
PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models
PriorVLA preserves pretrained priors in VLA models through a frozen Prior Expert and trained Adaptation Expert, delivering better robot manipulation performance than full fine-tuning with only 25% of the parameter updates.
-
Toward Visually Realistic Simulation: A Benchmark for Evaluating Robot Manipulation in Simulation
VISER is a new visually realistic simulation benchmark for robot manipulation tasks that uses PBR materials and MLLM-assisted asset generation, achieving 0.92 Pearson correlation with real-world policy performance.
-
MolmoAct2: Action Reasoning Models for Real-world Deployment
MolmoAct2 is an open VLA model that outperforms baselines like Pi-05 on 7 benchmarks and whose backbone surpasses GPT-5 on 13 embodied-reasoning tasks through new datasets, specialized training, and architecture chang...
-
Source-Modality Monitoring in Vision-Language Models
Vision-language models use semantic signals more than syntactic ones to bind words like 'image' to actual visual inputs, with implications for robustness in multimodal systems.
-
SafetyALFRED: Evaluating Safety-Conscious Planning of Multimodal Large Language Models
SafetyALFRED shows multimodal LLMs recognize kitchen hazards accurately in QA tests but achieve low success rates when required to mitigate those hazards through embodied planning.
-
If you're waiting for a sign... that might not be it! Mitigating Trust Boundary Confusion from Visual Injections on Vision-Language Agentic Systems
LVLM-based agents exhibit trust boundary confusion with visual injections and a multi-agent defense separating perception from decision-making reduces misleading responses while preserving correct ones.
-
FASTER: Rethinking Real-Time Flow VLAs
FASTER adds a Horizon-Aware Schedule to flow VLAs that compresses immediate-action denoising to one step while keeping long-horizon trajectory quality, lowering real-robot reaction latency.
-
FASTER: Rethinking Real-Time Flow VLAs
FASTER uses a horizon-aware flow sampling schedule to compress immediate-action denoising to one step, slashing effective reaction latency in real-robot VLA deployments.
-
RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies
RoboMME is a new benchmark with 16 tasks and 14 memory-augmented VLA variants that shows memory effectiveness is highly task-dependent.
-
Self-Supervised Bootstrapping of Action-Predictive Embodied Reasoning
R&B-EnCoRe uses self-supervised importance-weighted variational inference to distill action-predictive reasoning datasets that improve VLA performance on manipulation, navigation, and driving tasks without external verifiers.
-
CLAMP: Contrastive Learning for 3D Multi-View Action-Conditioned Robotic Manipulation Pretraining
CLAMP pretrains 3D multi-view encoders with contrastive learning on point clouds and actions, then initializes diffusion policies for more sample-efficient fine-tuning on robotic tasks.
-
A Pragmatic VLA Foundation Model
LingBot-VLA is a VLA foundation model trained on massive real robot data that shows superior generalization across tasks and platforms with fast training throughput.
-
FOCA: Future-Oriented Conditioning for Data-Efficient Vision-Language-Action Adaptation
FOCA improves few-shot VLA adaptation by explicitly predicting future interaction embeddings and implicitly aligning to goal observations, yielding up to 26% gains on real robots with only 20 demonstrations.
-
ZeroDex: Zero-Shot Long-Horizon Dexterous Manipulation via Multi-View 3D-Grounded VLM Reasoning
ZeroDex grounds VLM outputs into 3D keypoints via multi-view triangulation and ray voting to enable zero-shot long-horizon dexterous manipulation with closed-loop replanning.
-
Sword: Style-Robust World Models as Simulators via Dynamic Latent Bootstrapping for VLA Policy Post-Training
Sword improves world model simulators for VLA policies by disentangling visual style from dynamics and bootstrapping latents for better consistency, outperforming baselines on LIBERO in generalization and RL post-trai...
-
Cooptimizing Safety and Performance Using Safety Value-Constrained Model Predictive Control
Augments MPC with a safety value function terminal constraint to achieve recursive feasibility and persistent safety while co-optimizing performance.
-
RoboAgent: Chaining Basic Capabilities for Embodied Task Planning
RoboAgent chains basic vision-language capabilities inside a single VLM via a scheduler and trains it in three stages (behavior cloning, DAgger, RL) to improve embodied task planning.
-
Multi-Agent Robotic Control with Onboard Vision-Language Models
A multi-agent architecture using compact VLMs on edge hardware controls a mobile manipulator across five warehouse task categories in hardware-in-the-loop simulation.
-
Bridging Semantics and Kinematics: A Modular Framework for Zero-Shot Robotic Manipulation
A modular framework using FastSAM with Set-of-Mark prompting, an LLM as semantic router, and MoveIt Task Constructor achieves 62% end-to-end success in zero-shot language-guided robotic manipulation across open-world ...
-
EXPO-FT: Sample-Efficient Reinforcement Learning Finetuning for Vision-Language-Action Models
EXPO-FT enables pretrained VLA policies to reach 30/30 success on complex manipulation tasks using an average of 19.1 minutes of online robot data while outperforming prior RL approaches.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.