SARL optimizes language prompt inputs to generalist vision-language-action policies through online RL to solve complex long-horizon tasks by composing existing skills.
super hub Canonical reference
Gemini Robotics: Bringing AI into the Physical World
Canonical reference. 82% of citing Pith papers cite this work as background.
122
Pith papers citing it
5
external citations · Pith
Background
82% of classified citations
abstract
Recent advancements in large multimodal models have led to the emergence of remarkable generalist capabilities in digital domains, yet their translation to physical agents such as robots remains a significant challenge. This report introduces a new family of AI models purposefully designed for robotics and built upon the foundation of Gemini 2.0. We present Gemini Robotics, an advanced Vision-Language-Action (VLA) generalist model capable of directly controlling robots. Gemini Robotics executes smooth and reactive movements to tackle a wide range of complex manipulation tasks while also being robust to variations in object types and positions, handling unseen environments as well as following diverse, open vocabulary instructions. We show that with additional fine-tuning, Gemini Robotics can be specialized to new capabilities including solving long-horizon, highly dexterous tasks, learning new short-horizon tasks from as few as 100 demonstrations and adapting to completely novel robot embodiments. This is made possible because Gemini Robotics builds on top of the Gemini Robotics-ER model, the second model we introduce in this work. Gemini Robotics-ER (Embodied Reasoning) extends Gemini's multimodal reasoning capabilities into the physical world, with enhanced spatial and temporal understanding. This enables capabilities relevant to robotics including object detection, pointing, trajectory and grasp prediction, as well as multi-view correspondence and 3D bounding box predictions. We show how this novel combination can support a variety of robotics applications. We also discuss and address important safety considerations related to this new class of robotics foundation models. The Gemini Robotics family marks a substantial step towards developing general-purpose robots that realizes AI's potential in the physical world.
hub tools
citation-role summary
background 29
dataset 4
baseline 1
citation-polarity summary
claims ledger
- abstract Recent advancements in large multimodal models have led to the emergence of remarkable generalist capabilities in digital domains, yet their translation to physical agents such as robots remains a significant challenge. This report introduces a new family of AI models purposefully designed for robotics and built upon the foundation of Gemini 2.0. We present Gemini Robotics, an advanced Vision-Language-Action (VLA) generalist model capable of directly controlling robots. Gemini Robotics executes smooth and reactive movements to tackle a wide range of complex manipulation tasks while also being
authors
co-cited works
representative citing papers
VLA language backbones show high redundancy on manipulation benchmarks, with half the LLM blocks removable and even two blocks sufficient to recover baseline performance after fine-tuning, unlike vision and action pathways.
Flow Reversal Steering steers flow matching generalist policies by reversing suboptimal actions to nearby better modes, enabling improved zero-shot control, quick distillation, and RL bootstrapping in robotic 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.
Foresight uses iterative VLM plan proposal and critique with RL from human feedback to raise navigation success 37% and cut interventions 52% in real-world tests.
Ambient Diffusion Policy enables better imitation learning from suboptimal robot data by leveraging spectral properties to restrict data usage to specific diffusion times.
VLMs exhibit consistent vertical-distance entanglement in embeddings from perspective bias in natural images, producing accuracy gaps that a new synthetic benchmark SpatialTunnel exposes as model-intrinsic.
DriveSpatial benchmark shows the strongest of 15 VLMs trails humans by 28.4 points on spatiotemporal tasks, with cognitive scene construction as the primary weakness.
RIO introduces a lightweight open-source framework that abstracts real-time robot I/O to support easy switching between embodiments and platforms for collecting data and deploying VLAs.
GuardVLA embeds a stealthy backdoor watermark in VLAs via secret messages in visual data and uses a swap-and-detect mechanism for post-release ownership verification that preserves task performance.
Reducing visual input to one token per frame in VLA world models maintains or improves long-horizon performance on MetaWorld, LIBERO, and real-robot tasks.
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
π₀.₇ is a steerable generalist robotic model that uses rich multimodal prompts including language, subgoal images, and performance metadata to achieve out-of-the-box generalization across tasks and robot bodies.
ATI is a tripartite bio-inspired architecture for physical AI that co-designs sensing and inference, shown in a camera prototype to raise accuracy from 53.8% to 88% and cut remote invocations by 43.3%.
VP-VLA decouples high-level reasoning from low-level control in VLA models by rendering spatial anchors as visual prompts directly in the RGB observation space, outperforming end-to-end baselines.
A video foundation model trained on human demonstrations generates zero-shot plans that convert to executable robot actions on novel scenes and tasks.
DreamGen trains robot policies on synthetic trajectories from adapted video world models, enabling a humanoid robot to perform 22 new behaviors in seen and unseen environments from a single pick-and-place teleoperation dataset.
A memory-guided LLM harness that calls a frozen VLA only for contact-rich phases lifts success to 82.4% on LIBERO-Pro, 55.4% on RoboCasa365, and 58.4% on RoboTwin C2R with no policy finetuning.
Temporal Ratio, the action head’s attention to future video latents versus the current frame, predicts and can be used to mitigate the compositional video–action generalization gap.
ROSA introduces shared GPU-pool serving, robotics-aware abstractions for multi-model pipelines, and factory-productivity scheduling that improves output by up to 12.06x over dedicated per-robot systems.
A plug-in Q-Synth Module plus Attention-to-Point Decoder converts text-conditioned attention in frozen MLLMs into point heatmaps, improving part-level grounding accuracy on multiple datasets.
CoStream composes semantic, predictive, and reactive behaviors on an SE(3) interface to enable precise, generalizable performance on eight real-world contact-rich manipulation tasks.
GRA extracts 2D waypoints from synthetic videos to supervise VLA vision while restricting action training to real data, outperforming pseudo-action baselines on real-robot tasks.
citing papers explorer
-
Adapting Generalist Robot Policies with Semantic Reinforcement Learning
SARL optimizes language prompt inputs to generalist vision-language-action policies through online RL to solve complex long-horizon tasks by composing existing skills.
-
Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?
VLA language backbones show high redundancy on manipulation benchmarks, with half the LLM blocks removable and even two blocks sufficient to recover baseline performance after fine-tuning, unlike vision and action pathways.
-
Improving Robotic Generalist Policies via Flow Reversal Steering
Flow Reversal Steering steers flow matching generalist policies by reversing suboptimal actions to nearby better modes, enabling improved zero-shot control, quick distillation, and RL bootstrapping in robotic manipulation.
-
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.
-
Foresight: Iterative Reasoning About Clues that Matter for Navigation
Foresight uses iterative VLM plan proposal and critique with RL from human feedback to raise navigation success 37% and cut interventions 52% in real-world tests.
-
Ambient Diffusion Policy: Imitation Learning from Suboptimal Data in Robotics
Ambient Diffusion Policy enables better imitation learning from suboptimal robot data by leveraging spectral properties to restrict data usage to specific diffusion times.
-
Why Far Looks Up: Probing Spatial Representation in Vision-Language Models
VLMs exhibit consistent vertical-distance entanglement in embeddings from perspective bias in natural images, producing accuracy gaps that a new synthetic benchmark SpatialTunnel exposes as model-intrinsic.
-
DRIVESPATIAL: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving
DriveSpatial benchmark shows the strongest of 15 VLMs trails humans by 28.4 points on spatiotemporal tasks, with cognitive scene construction as the primary weakness.
-
RIO: Flexible Real-Time Robot I/O for Cross-Embodiment Robot Learning
RIO introduces a lightweight open-source framework that abstracts real-time robot I/O to support easy switching between embodiments and platforms for collecting data and deploying VLAs.
-
Towards Backdoor-Based Ownership Verification for Vision-Language-Action Models
GuardVLA embeds a stealthy backdoor watermark in VLAs via secret messages in visual data and uses a swap-and-detect mechanism for post-release ownership verification that preserves task performance.
-
One Token Per Frame: Reconsidering Visual Bandwidth in World Models for VLA Policy
Reducing visual input to one token per frame in VLA world models maintains or improves long-horizon performance on MetaWorld, LIBERO, and real-robot tasks.
-
NoiseGate: Learning Per-Latent Timestep Schedules as Information Gating in World Action Models
NoiseGate learns per-latent timestep schedules as an information-gating policy in diffusion-based world action models, yielding consistent gains on RoboTwin manipulation tasks.
-
Being-H0.7: A Latent World-Action Model from Egocentric Videos
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
-
${\pi}_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities
π₀.₇ is a steerable generalist robotic model that uses rich multimodal prompts including language, subgoal images, and performance metadata to achieve out-of-the-box generalization across tasks and robot bodies.
-
[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI
ATI is a tripartite bio-inspired architecture for physical AI that co-designs sensing and inference, shown in a camera prototype to raise accuracy from 53.8% to 88% and cut remote invocations by 43.3%.
-
VP-VLA: Visual Prompting as an Interface for Vision-Language-Action Models
VP-VLA decouples high-level reasoning from low-level control in VLA models by rendering spatial anchors as visual prompts directly in the RGB observation space, outperforming end-to-end baselines.
-
Large Video Planner Enables Generalizable Robot Control
A video foundation model trained on human demonstrations generates zero-shot plans that convert to executable robot actions on novel scenes and tasks.
-
DreamGen: Unlocking Generalization in Robot Learning through Video World Models
DreamGen trains robot policies on synthetic trajectories from adapted video world models, enabling a humanoid robot to perform 22 new behaviors in seen and unseen environments from a single pick-and-place teleoperation dataset.
-
Harness VLA: Steering Frozen VLAs into Reliable Manipulation Primitives via Memory-Guided Agents
A memory-guided LLM harness that calls a frozen VLA only for contact-rich phases lifts success to 82.4% on LIBERO-Pro, 55.4% on RoboCasa365, and 58.4% on RoboTwin C2R with no policy finetuning.
-
Understanding and Mitigating the Video-Action Generalization Gap via Temporal Ratio
Temporal Ratio, the action head’s attention to future video latents versus the current frame, predicts and can be used to mitigate the compositional video–action generalization gap.
-
ROSA: A Robotics Foundation Model Serving System for Robot Factories
ROSA introduces shared GPU-pool serving, robotics-aware abstractions for multi-model pipelines, and factory-productivity scheduling that improves output by up to 12.06x over dedicated per-robot systems.
-
Enhancing Part-Level Point Grounding for Any Open-Source MLLMs
A plug-in Q-Synth Module plus Attention-to-Point Decoder converts text-conditioned attention in frozen MLLMs into point heatmaps, improving part-level grounding accuracy on multiple datasets.
-
CoStream: Composing Simple Behaviors for Generalizable Complex Manipulation
CoStream composes semantic, predictive, and reactive behaviors on an SE(3) interface to enable precise, generalizable performance on eight real-world contact-rich manipulation tasks.
-
Supervise What Survives: Geometry-Guided VLA Adaptation from Synthetic Robot Videos
GRA extracts 2D waypoints from synthetic videos to supervise VLA vision while restricting action training to real data, outperforming pseudo-action baselines on real-robot tasks.
-
A Watermark for Vision-Language-Action and World Action Models
A method to fingerprint VLA and WAM policies by keyed Gaussian noise seeds enables reliable ownership verification from partial outputs with minimal performance impact.
-
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%.
-
DREAM-Chunk: Reactive Action Chunking with Latent World Model
DREAM-Chunk uses test-time sampling and latent-world-model rollouts to select robust action chunks from chunking-based VLA policies, improving performance under stochastic dynamics on simulation and hardware tasks.
-
SERF: Spatiotemporal Environment and Robot Feature Map for Long-Horizon Mobile Manipulation
SERF conditions VLA policies on online-updated neural point maps of environment and robot to improve long-horizon mobile manipulation on BEHAVIOR-1K.
-
APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies
APT pretrains the action expert as a vision-action prior on frozen VLM features then adds language through gated fusion to improve OOD instruction generalization in continuous-action VLA policies.
-
Ouroboros-Spatial: Closing the Data-Model Loop for Spatial Reasoning
A self-evolving training loop that generates its own spatial QA data with executable code and difficulty feedback lifts Qwen3-VL-4B/8B to 62.7/63.3 on VSI-Bench using an order of magnitude less data.
-
Next Forcing: Causal World Modeling with Multi-Chunk Prediction
Next Forcing augments video generation models with auxiliary multi-chunk prediction modules to achieve faster training convergence, higher accuracy at high frame rates, and 2x faster inference on world modeling benchmarks.
-
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.
-
When Video Misreads: Closed-Loop Distillation of Reading Heuristics for Exploratory Manipulation Trace QA
Closed-Loop Trace Distillation distills one-line natural-language prompts from labeled training traces to improve VLM accuracy on predicting minimal-success action chains in Exploratory Manipulation Trace QA by 0.38-0.47 across simulator and real-robot tasks.
-
AffordanceVLA: A Vision-Language-Action Model Empowering Action Generation through Affordance-Aware Understanding
AffordanceVLA proposes a VLA model with affordance-aware modules (Which2Act, Where2Act, How2Act) in a Mixture-of-Transformer trained in three stages to improve robotic manipulation.
-
Probing Collision Grounding in Vision-Language Models for Safe Human-Robot Collaboration
TouchSafeBench evaluates VLMs on collision grounding, finding best Macro-F1 below 50% and that explicit depth does not yield reliable robot-body contact inference.
-
Task-Focused Memorization for Multimodal Agents
TaskMem uses RL in two phases to learn a task-focused memorization policy for multimodal agents, yielding 5.3-7.0% VQA accuracy gains on reformulated streaming benchmarks from VideoMME, EgoLife, and EgoTempo.
-
Goal-Conditioned Agents that Learn Everything All at Once
LEO enables efficient all-goals learning in goal-conditioned RL by jointly predicting for all goals in one network pass, yielding >250x speedup over relabelling and better performance on Craftax.
-
Cambrian-P: Pose-Grounded Video Understanding
Adding per-frame camera-pose supervision to a video MLLM improves spatial and general video question answering by 2–6% and yields SOTA streaming pose estimates on ScanNet.
-
Action with Visual Primitives
AVP architecture has VLM emit visual-primitive tokens to condition flow-matching action expert, yielding 27.61% higher success rate than pi_0.5 on real-robot pick-and-place tasks.
-
From a Single Demonstration to a General Policy for Contact-Rich Manipulation
A one-shot LfD framework abstracts a single demonstration into environmental-constraint primitives, then uses self-exploration, human corrections, and compliant recovery to produce a policy that generalizes across poses and geometries, achieving over 90% success on seven real-world multi-stage tasks
-
GuidedVLA: Specifying Task-Relevant Factors via Plug-and-Play Action Attention Specialization
GuidedVLA improves VLA generalization by supervising individual attention heads with manually defined auxiliary signals for three task-relevant factors.
-
Nautilus: From One Prompt to Plug-and-Play Robot Learning
A typed-contract harness with containerized 'chambers' and robotics-specific agent skills lets a coding LLM turn a single natural-language prompt into working reproduction, evaluation, and deployment workflows for robot learning.
-
Beyond Thinking: Imagining in 360$^\circ$ for Humanoid Visual Search
Imagining in 360° decouples visual search into a single-step probabilistic semantic layout predictor and an actor, removing the need for multi-turn CoT reasoning and trajectory annotations while improving efficiency in 360° environments.
-
Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation
Anchor-Centric Adaptation escapes the diversity trap by prioritizing repeated demonstrations at core anchors over broad coverage, yielding higher success rates under fixed data budgets in robotic manipulation.
-
Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning
Odysseus adapts PPO with a turn-level critic and leverages pretrained VLM action priors to train agents achieving at least 3x average game progress over frontier models in long-horizon Super Mario Land.
-
Breaking Lock-In: Preserving Steerability under Low-Data VLA Post-Training
DeLock mitigates lock-in in low-data VLA post-training via visual grounding preservation and test-time contrastive prompt guidance, outperforming baselines across eight evaluations while matching data-heavy generalist policies.
-
RL Token: Bootstrapping Online RL with Vision-Language-Action Models
RL Token enables sample-efficient online RL fine-tuning of large VLAs, delivering up to 3x speed gains and higher success rates on real-robot manipulation tasks within minutes to hours.
-
GazeVLA: Learning Human Intention for Robotic Manipulation
GazeVLA pretrains on large human egocentric datasets to capture gaze-based intention, then finetunes on limited robot data with chain-of-thought reasoning to achieve better robotic manipulation performance than baselines.
-
Learning-augmented robotic automation for real-world manufacturing
A learning-augmented robotic system automated deformable cable insertion and soldering on a live electric-motor production line for 5 hours 10 minutes, producing 108 motors at 99.4% pass rate with under 20 minutes of real-world data per task and no physical fencing.
-
dWorldEval: Scalable Robotic Policy Evaluation via Discrete Diffusion World Model
A discrete diffusion model tokenizes multimodal robotic data and uses a progress token to predict future states and task completion for scalable policy evaluation.