AgentDojo introduces an extensible evaluation framework populated with realistic agent tasks and security test cases to measure prompt injection robustness in tool-using LLM agents.
hub Canonical reference
A Generalist Agent
Canonical reference. 94% of citing Pith papers cite this work as background.
62
Pith papers citing it
Background
94% of classified citations
abstract
Inspired by progress in large-scale language modeling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment generalist policy. The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens. In this report we describe the model and the data, and document the current capabilities of Gato.
hub tools
citation-role summary
background 16
baseline 1
citation-polarity summary
representative citing papers
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
A hypernetwork generates complete task-specific visuomotor policy parameters from instructions alone to structurally eliminate observation leakage in language-conditioned robotic control.
Aero-World adapts a pretrained latent diffusion transformer for action-conditioned aerial video generation by injecting inertial action tokens and using a frozen latent-space Physics Probe for inertial consistency supervision during LoRA finetuning, with a new AeroBench benchmark showing improved AA
TokaMind, pre-trained on MAST tokamak data, transfers to power grid PMU data for severe event classification with F1 0.837, where difficulty depends on grid topology and CSD indicators boost early-warning performance over CNN baselines.
Multi-layer transformers can implement in-context logistic regression by performing normalized gradient descent steps layer by layer, obtained via supervised training of a single attention layer followed by recurrent application with convergence and OOD guarantees.
KITE is a training-free method that uses keyframe-indexed tokenized evidence including BEV schematics to enhance VLM performance on robot failure detection, identification, localization, explanation, and correction.
Factorization Regret measures how latent variable interactions affect performance, and RCCs enable learning them to achieve compositional generalization in partially observable tasks.
Closed-loop prompt-based translation with hierarchical verification and iterative repair produces equivalent high-performance RL environments across five cases including new TCGJax.
ST-BiBench reveals a coordination paradox in which MLLMs show strong high-level strategic reasoning yet fail at fine-grained 16-dimensional bimanual action synthesis and multi-stream fusion.
A survey that defines Compound AI Systems, proposes a multi-dimensional taxonomy based on component roles and orchestration strategies, reviews four foundational paradigms, and identifies key challenges for future research.
ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.
UniSim learns a universal real-world simulator from orchestrated diverse datasets, enabling zero-shot deployment of policies trained purely in simulation.
DreamerV3 uses world models and robustness techniques to solve over 150 tasks across domains with a single configuration, including Minecraft diamond collection from scratch.
pcsp is a shared RL policy using LLM persona embeddings, low-rank projection, and PPO+InfoNCE+KL training that delivers 17x above-chance zero-shot persona identification and 22x faster inference on a 300-persona benchmark.
DiLA uses content-structure disentanglement driven by predictive bottlenecks to create semantically structured latent actions for high-fidelity video world models.
Geometric Pareto Control embeds Pareto solutions in a Lie group submanifold and navigates via Riemannian gradient flow to achieve 100% feasibility and low suboptimality in control tasks without retraining.
RELO formulates visual object tracking localization as a Markov decision process solved by reinforcement learning with combined IoU and AUC rewards, augmented by layer-aligned temporal token propagation, and reports 57.5% AUC on LaSOText without template updates.
MetaRL pre-trained on GBWM problems delivers near-optimal dynamic strategies in 0.01s achieving 97.8% of DP optimal utility and handles larger problems where DP fails.
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.
M²-VLA shows that generalized VLMs can serve as direct backbones for robotic manipulation by selectively extracting task-critical features via Mixture of Layers and adding Meta Skill Modules for efficient trajectory learning.
CMAT uses a transformer decoder to produce a high-level consensus vector in latent space, enabling simultaneous order-independent actions by all agents and optimization via single-agent PPO, with superior results on StarCraft II, Multi-Agent MuJoCo, and Google Research Football.
ProGAL-VLA uses 3D graphs, symbolic sub-goals, and a Grounding Alignment Contrastive loss to ground actions on verified embeddings, raising robustness from 30.3% to 71.5% and ambiguity AUROC to 0.81 on robotic benchmarks.
ADAPT is a new pre-training paradigm that aligns physical properties of time-series data to allow simultaneous training on 162 diverse classification datasets, achieving new state-of-the-art performance.
citing papers explorer
-
AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
AgentDojo introduces an extensible evaluation framework populated with realistic agent tasks and security test cases to measure prompt injection robustness in tool-using LLM agents.
-
Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
-
DISC: Decoupling Instruction from State-Conditioned Control via Policy Generation
A hypernetwork generates complete task-specific visuomotor policy parameters from instructions alone to structurally eliminate observation leakage in language-conditioned robotic control.
-
Aero-World: Action-Conditioned Aerial Video Generation from Inertial Controls
Aero-World adapts a pretrained latent diffusion transformer for action-conditioned aerial video generation by injecting inertial action tokens and using a frozen latent-space Physics Probe for inertial consistency supervision during LoRA finetuning, with a new AeroBench benchmark showing improved AA
-
TokaMind for Power Grid: Cross-Domain Transfer from Fusion Plasma
TokaMind, pre-trained on MAST tokamak data, transfers to power grid PMU data for severe event classification with F1 0.837, where difficulty depends on grid topology and CSD indicators boost early-warning performance over CNN baselines.
-
Transformers Efficiently Perform In-Context Logistic Regression via Normalized Gradient Descent
Multi-layer transformers can implement in-context logistic regression by performing normalized gradient descent steps layer by layer, obtained via supervised training of a single attention layer followed by recurrent application with convergence and OOD guarantees.
-
KITE: Keyframe-Indexed Tokenized Evidence for VLM-Based Robot Failure Analysis
KITE is a training-free method that uses keyframe-indexed tokenized evidence including BEV schematics to enhance VLM performance on robot failure detection, identification, localization, explanation, and correction.
-
Factorization Regret mediates compositional generalization in latent space
Factorization Regret measures how latent variable interactions affect performance, and RCCs enable learning them to achieve compositional generalization in partially observable tasks.
-
Automatic Generation of High-Performance RL Environments
Closed-loop prompt-based translation with hierarchical verification and iterative repair produces equivalent high-performance RL environments across five cases including new TCGJax.
-
ST-BiBench: Benchmarking Multi-Stream Multimodal Coordination in Bimanual Embodied Tasks for MLLMs
ST-BiBench reveals a coordination paradox in which MLLMs show strong high-level strategic reasoning yet fail at fine-grained 16-dimensional bimanual action synthesis and multi-stream fusion.
-
From Standalone LLMs to Integrated Intelligence: A Survey of Compound Al Systems
A survey that defines Compound AI Systems, proposes a multi-dimensional taxonomy based on component roles and orchestration strategies, reviews four foundational paradigms, and identifies key challenges for future research.
-
Any-point Trajectory Modeling for Policy Learning
ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.
-
Learning Interactive Real-World Simulators
UniSim learns a universal real-world simulator from orchestrated diverse datasets, enabling zero-shot deployment of policies trained purely in simulation.
-
Mastering Diverse Domains through World Models
DreamerV3 uses world models and robustness techniques to solve over 150 tasks across domains with a single configuration, including Minecraft diamond collection from scratch.
-
One Policy, Infinite NPCs: Persona-Traceable Shared RL Policies for Scalable Game Agents
pcsp is a shared RL policy using LLM persona embeddings, low-rank projection, and PPO+InfoNCE+KL training that delivers 17x above-chance zero-shot persona identification and 22x faster inference on a 300-persona benchmark.
-
DiLA: Disentangled Latent Action World Models
DiLA uses content-structure disentanglement driven by predictive bottlenecks to create semantically structured latent actions for high-fidelity video world models.
-
Geometric Pareto Control: Riemannian Gradient Flow of Energy Function via Lie Group Homotopy
Geometric Pareto Control embeds Pareto solutions in a Lie group submanifold and navigates via Riemannian gradient flow to achieve 100% feasibility and low suboptimality in control tasks without retraining.
-
RELO: Reinforcement Learning to Localize for Visual Object Tracking
RELO formulates visual object tracking localization as a Markov decision process solved by reinforcement learning with combined IoU and AUC rewards, augmented by layer-aligned temporal token propagation, and reports 57.5% AUC on LaSOText without template updates.
-
A Meta Reinforcement Learning Approach to Goals-Based Wealth Management
MetaRL pre-trained on GBWM problems delivers near-optimal dynamic strategies in 0.01s achieving 97.8% of DP optimal utility and handles larger problems where DP fails.
-
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.
-
$M^2$-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills
M²-VLA shows that generalized VLMs can serve as direct backbones for robotic manipulation by selectively extracting task-critical features via Mixture of Layers and adding Meta Skill Modules for efficient trajectory learning.
-
Bridging MARL to SARL: An Order-Independent Multi-Agent Transformer via Latent Consensus
CMAT uses a transformer decoder to produce a high-level consensus vector in latent space, enabling simultaneous order-independent actions by all agents and optimization via single-agent PPO, with superior results on StarCraft II, Multi-Agent MuJoCo, and Google Research Football.
-
ProGAL-VLA: Grounded Alignment through Prospective Reasoning in Vision-Language-Action Models
ProGAL-VLA uses 3D graphs, symbolic sub-goals, and a Grounding Alignment Contrastive loss to ground actions on verified embeddings, raising robustness from 30.3% to 71.5% and ambiguity AUROC to 0.81 on robotic benchmarks.
-
ADAPTive Input Training for Many-to-One Pre-Training on Time-Series Classification
ADAPT is a new pre-training paradigm that aligns physical properties of time-series data to allow simultaneous training on 162 diverse classification datasets, achieving new state-of-the-art performance.
-
Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
Evo-Memory is a new streaming benchmark and evaluation framework for self-evolving memory in LLM agents, unifying over ten memory modules and introducing the ReMem pipeline for continual improvement on multi-turn and reasoning datasets.
-
AsyncVLA: Asynchronous Flow Matching for Vision-Language-Action Models
AsyncVLA adds asynchronous flow matching and a confidence rater to VLA models so they can generate actions on flexible schedules and selectively refine low-confidence tokens before execution.
-
DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge
DreamVLA uses dynamic-region-guided world knowledge prediction, block-wise attention to disentangle information types, and a diffusion transformer for actions, reaching 76.7% success on real robot tasks and 4.44 average length on CALVIN ABC-D.
-
Preference Goal Tuning: Post-Training as Latent Control for Frozen Policies
PGT optimizes latent goal embeddings for frozen policies via trajectory-level preference objectives, reporting 72-81.6% relative gains on 17 Minecraft tasks and 13.4% better OOD performance than fine-tuning.
-
DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning
DINO-WM builds world models on pre-trained DINOv2 features to enable zero-shot planning from offline data without rewards or demonstrations.
-
GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation
GR-2 pre-trains on web-scale videos then fine-tunes on robot data to reach 97.7% average success across over 100 manipulation tasks with strong generalization to new scenes and objects.
-
LLaVA-Video: Video Instruction Tuning With Synthetic Data
LLaVA-Video-178K is a new synthetic video instruction dataset that, when combined with existing data to train LLaVA-Video, produces strong results on video understanding benchmarks.
-
Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
-
3D Diffuser Actor: Policy Diffusion with 3D Scene Representations
3D Diffuser Actor unifies diffusion policies with 3D scene features to set new state-of-the-art results on RLBench and CALVIN robot benchmarks.
-
Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation
A GPT-style model pre-trained on large video datasets achieves 94.9% success on CALVIN multi-task manipulation and 85.4% zero-shot generalization, outperforming prior baselines.
-
Gemini: A Family of Highly Capable Multimodal Models
Gemini Ultra reaches human-expert performance on MMLU for the first time and sets new state-of-the-art results on 30 of 32 benchmarks, including all 20 multimodal ones tested.
-
TD-MPC2: Scalable, Robust World Models for Continuous Control
TD-MPC2 scales an implicit world-model RL method to a 317M-parameter agent that masters 80 tasks across four domains with a single hyperparameter configuration.
-
PaLM-E: An Embodied Multimodal Language Model
PaLM-E is a single 562B-parameter multimodal model that performs embodied reasoning tasks like robotic manipulation planning and visual question answering by interleaving vision, state, and text inputs with positive transfer from joint training on language and robotics data.
-
Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents
DEPS combines LLM-based interactive planning with a trainable goal selector to create a zero-shot multi-task agent that completes 70+ Minecraft tasks and nearly doubles prior performance.
-
Self-Evolving Cognitive Framework via Causal World Modeling for Embodied Scientific Intelligence
Proposes a self-evolving cognitive framework integrating causal world modeling, intervention-driven reasoning, and continual refinement for embodied scientific intelligence.
-
SAFE-Pruner: Semantic Attention-Guided Future-Aware Token Pruning for Efficient Vision-Language-Action Manipulation
SAFE-Pruner forecasts deep-layer token saliency in VLA models via semantic attention consistency and adaptive subtask detection to achieve up to 1.89x speedup with under 1.7% success rate loss.
-
SWEET: Sparse World Modeling with Image Editing for Embodied Task Execution
SWEET is a one-shot sparse visual planning framework that progressively generates manipulation keyframes via image editing conditioned on language and spatial guidance, then converts them to actions with a diffusion predictor, showing better fidelity and lower cost than video models on DROID and Rob
-
Large Language Models for Sequential Decision-Making: Improving In-Context Learning via Supervised Fine-Tuning
Supervised fine-tuning of pretrained LLMs on offline trajectories yields better few-shot sequential decision-making than in-context-only baselines, with a theoretical suboptimality bound derived for linear MDPs by interpreting attention as Q-function estimation.
-
Towards Systematic Generalization for Power Grid Optimization Problems
A shared graph neural network framework jointly solves ACOPF and SCUC problems using physics constraints and shows improved generalization to unseen grid topologies.
-
Neural Computers
Neural Computers are introduced as a new machine form where computation, memory, and I/O are unified in a learned runtime state, with initial video-model experiments showing acquisition of basic interface primitives from traces.
-
A Hierarchical Error-Corrective Graph Framework for Autonomous Agents with LLM-Based Action Generation
HECG combines multi-dimensional metrics for strategy choice, ten-type error classification with recoverability details, and causal-context graphs to improve LLM agent reliability in complex tasks.
-
UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning
UI-TARS-2 reaches 88.2 on Online-Mind2Web, 47.5 on OSWorld, 50.6 on WindowsAgentArena, and 73.3 on AndroidWorld while attaining 59.8 mean normalized score on a 15-game suite through multi-turn RL and scalable data generation.
-
GR-3 Technical Report
GR-3 is a VLA model that generalizes to novel objects, environments, and abstract instructions, outperforms the π0 baseline, and integrates with the new ByteMini bi-manual mobile robot.
-
SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning
SafeVLA applies constrained reinforcement learning via CMDP min-max optimization to VLAs, cutting safety violation costs by 83.58% while preserving task success on long-horizon mobile manipulation tasks.
-
What Matters in Building Vision-Language-Action Models for Generalist Robots
Systematic tests of VLM backbones, policy architectures, and cross-embodiment data yield RoboVLMs that set new SOTA on robot manipulation benchmarks while requiring few manual designs.
-
Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks
Magentic-One is a modular multi-agent system that matches state-of-the-art performance on GAIA, AssistantBench, and WebArena using an orchestrator-led team of specialized agents.