Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long-horizon, economically valuable, real-world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 subfields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is 2.6%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP-relevant impact.

Affordance2Action: Task-Conditioned Scene-level Affordance Grounding for Real-Time Manipulation

Task-conditioned manipulation requires grounding instructions to task-relevant functional parts rather than object categories. This setting is scene-dependent and often one-to-many in cluttered scenes: the same object may afford different interactions across tasks, while a single task may correspond to either one functional region or multiple valid functional regions, depending on the scene layout. Existing affordance datasets and benchmarks remain misaligned with this setting, as they typically focus on grasping or object-level affordances, rely on synthetic scenes, or assume a single instruction-region correspondence. We present Affordance2Action (A2A), a benchmark-centered learning framework for scene-level, task-conditioned part affordance grounding. At its core is A2A-Bench, a manipulation-oriented benchmark that covers both single-region and multi-region instruction correspondences in everyday scenes, with the latter highlighting the ambiguity and diversity of affordance grounding in realistic multi-object environments. To construct it at scale, we build A2A-AffordGen, an agent-assisted annotation pipeline that combines language-model filtering, interactive part segmentation, instance-level mask-out refinement, task-reasoning instruction generation, and human verification. A2A-Bench's supervision further supports diverse downstream applications, with real-time affordance grounding and affordance-conditioned manipulation policies as two representative examples. Experiments show that A2A exposes substantial gaps in generic segmentation, VLM-based grounding, and affordance distillation baselines, while improving task-level localization and providing useful spatial priors for downstream manipulation. All datasets and code will be publicly released to promote open research.

BORA: Bridging Offline Reinforcement Learning and Online Residual Adaptation for Real-World Dexterous VLA Models

Vision-Language-Action (VLA) models have emerged as a promising paradigm for grounding visual-language understanding into real-world robotic manipulation. However, dexterous manipulation remains challenging for VLA policies due to high-dimensional hand control and compounding execution errors, which makes real-world RL post-training essential for bridging the gap between visually grounded action generation and physically reliable dexterous execution. However, high-dimensional dexterous exploration often triggers temporal inconsistency, sample inefficiency and hardware risks in the real world. To address these challenges, we propose BORA, an offline-to-online RL post-training framework designed for real-world dexterous VLA models. In the offline phase, BORA constructs a critic that takes both the VLM's cognition tokens and action chunks as inputs. This design enables action-conditioned value guidance, allowing the critic to evaluate dexterous hand motions beyond visual context alone. During the subsequent online phase, BORA freezes the VLA base and introduces a lightweight, Human-in-the-Loop (HiL) chunk-wise residual adaptation mechanism to mitigate real-world execution errors and further correct the offline-learned intents within the actual physical environment. By inheriting the offline critic and employing intervention-driven rewards, BORA effectively corrects execution discrepancies and adapts to real-world physical variances while preserving the pretrained policy as a stable prior. Extensive evaluations across five complex real-world dexterous tasks demonstrate that BORA significantly outperforms pure imitation learning and traditional decoupled RL baselines, achieving a 33% absolute increase in average success rate under standard settings and up to a 43% improvement in unseen object generalization.

BridgeACT: Bridging Human Demonstrations to Robot Actions via Unified Tool-Target Affordances

Learning robot manipulation from human videos is appealing due to the scale and diversity of human demonstrations, but transferring such demonstrations to executable robot behavior remains challenging. Prior work either relies on robot data for downstream adaptation or learns affordance representations that remain at the perception level and do not directly support real-world execution. We present BridgeACT, an affordance-driven framework that learns robotic manipulation directly from human videos without requiring any robot demonstration data. Our key idea is to model affordance as an embodiment-agnostic intermediate representation that bridges human demonstrations and robot actions. BridgeACT decomposes manipulation into two complementary problems: where to grasp and how to move. To this end, BridgeACT first grounds task-relevant affordance regions in the current scene, and then predicts task-conditioned 3D motion affordances from human demonstrations. The resulting affordances are mapped to robot actions through a grasping module and a lightweight closed-loop motion controller, enabling direct deployment on real robots. In addition, we represent complex manipulation tasks as compositions of affordance operations, which allows a unified treatment of diverse tasks and object-to-object interactions. Experiments on real-world manipulation tasks show that BridgeACT outperforms prior baselines and generalizes to unseen objects, scenes, and viewpoints.

DexHiL: A Human-in-the-Loop Framework for Vision-Language-Action Model Post-Training in Dexterous Manipulation

While Vision-Language-Action (VLA) models have demonstrated promising generalization capabilities in robotic manipulation, deploying them on specific and complex downstream tasks still demands effective post-training. In parallel, Human-in-the-Loop (HiL) learning has proven to be a powerful mechanism for refining robot policies. However, extending this paradigm to dexterous manipulation remains challenging: multi-finger control is high-dimensional, contact-intensive, and exhibits execution distributions that differ markedly from standard arm motions, leaving existing dexterous VLA systems limited in reliability and adaptability. We present DexHiL, the first integrated arm-hand human-in-the-loop framework for dexterous VLA models, enabling coordinated interventions over the arm and the dexterous hand within a single system. DexHiL introduces an intervention-aware data sampling strategy that prioritizes corrective segments for post-training, alongside a lightweight teleoperation interface that supports instantaneous human corrections during execution. Real-robot experiments demonstrate that DexHiL serves as an effective post-training framework, yielding a substantial performance leap, outperforming standard offline-only fine-tuning baselines by an average of 25% in success rates across distinct tasks. Project page: https://chenzhongxi-sjtu.github.io/dexhil/

Viewpoint Matters: Dynamically Optimizing Viewpoints with Masked Autoencoder for Visual Manipulation

Robotic manipulation continues to be a challenge, and imitation learning (IL) enables robots to learn tasks from expert demonstrations. Current IL methods typically rely on fixed camera setups, where cameras are manually positioned in static locations, imposing significant limitations on adaptability and coverage. Inspired by human active perception, where humans dynamically adjust their viewpoint to capture the most relevant and least noisy information, we propose MAE-Select, a novel framework for active viewpoint selection in single-camera robotic systems. MAE-Select fully leverages pre-trained multi-view masked autoencoder representations and dynamically selects the next most informative viewpoint at each time chunk without requiring labeled viewpoints. Extensive experiments demonstrate that MAE-Select improves the capabilities of single-camera systems and, in some cases, even surpasses multi-camera setups. The project will be available at https://mae-select.github.io.

FSAG: Enhancing Human-to-Dexterous-Hand Finger-Specific Affordance Grounding via Diffusion Models

Dexterous grasp synthesis remains a central challenge: the high dimensionality and kinematic diversity of multi-fingered hands prevent direct transfer of algorithms developed for parallel-jaw grippers. Existing approaches typically depend on large, hardware-specific grasp datasets collected in simulation or through costly real-world trials, hindering scalability as new dexterous hand designs emerge. To this end, we propose a data-efficient framework, which is designed to bypass robot grasp data collection by exploiting the rich, object-centric semantic priors latent in pretrained generative diffusion models. Temporally aligned and fine-grained grasp affordances are extracted from raw human video demonstrations and fused with 3D scene geometry from depth images to infer semantically grounded contact targets. A kinematics-aware retargeting module then maps these affordance representations to diverse dexterous hands without per-hand retraining. The resulting system produces stable, functionally appropriate multi-contact grasps that remain reliably successful across common objects and tools, while exhibiting strong generalization across previously unseen object instances within a category, pose variations, and multiple hand embodiments. This work (i) introduces a semantic affordance extraction pipeline leveraging vision-language generative priors for dexterous grasping, (ii) demonstrates cross-hand generalization without constructing hardware-specific grasp datasets, and (iii) establishes that a single depth modality suffices for high-performance grasp synthesis when coupled with foundation-model semantics. Our results highlight a path toward scalable, hardware-agnostic dexterous manipulation driven by human demonstrations and pretrained generative models.

Look, Zoom, Understand: The Robotic Eyeball for Embodied Perception

In embodied AI perception systems, visual perception should be active: the goal is not to passively process static images, but to actively acquire more informative data within pixel and spatial budget constraints. Existing vision models and fixed RGB-D camera systems fundamentally fail to reconcile wide-area coverage with fine-grained detail acquisition, severely limiting their efficacy in open-world robotic applications. To address this issue, we propose EyeVLA, a robotic eyeball for active visual perception that can take proactive actions based on instructions, enabling clear observation of fine-grained target objects and detailed information across a wide spatial extent. EyeVLA discretizes action behaviors into action tokens and integrates them with vision-language models (VLMs) that possess strong open-world understanding capabilities, enabling joint modeling of vision, language, and actions within a single autoregressive sequence. By using the 2D bounding box coordinates to guide the reasoning chain and applying reinforcement learning to refine the viewpoint selection policy, we transfer the open-world scene understanding capability of the VLM to a vision language action (VLA) policy using only minimal real-world data. Experiments show that our system efficiently performs instructed scenes in real-world environments and actively acquires more accurate visual information through instruction-driven actions of rotation and zoom, thereby achieving strong environmental perception capabilities. EyeVLA introduces a novel robotic vision system that leverages detailed and spatially rich, large-scale embodied data, and actively acquires highly informative visual observations for downstream embodied tasks.

SAGE: Scene Graph-Aware Guidance and Execution for Long-Horizon Manipulation Tasks

Successfully solving long-horizon manipulation tasks remains a fundamental challenge. These tasks involve extended action sequences and complex object interactions, presenting a critical gap between high-level symbolic planning and low-level continuous control. To bridge this gap, two essential capabilities are required: robust long-horizon task planning and effective goal-conditioned manipulation. Existing task planning methods, including traditional and LLM-based approaches, often exhibit limited generalization or sparse semantic reasoning. Meanwhile, image-conditioned control methods struggle to adapt to unseen tasks. To tackle these problems, we propose SAGE, a novel framework for Scene Graph-Aware Guidance and Execution in Long-Horizon Manipulation Tasks. SAGE utilizes semantic scene graphs as a structural representation for scene states. A structural scene graph enables bridging task-level semantic reasoning and pixel-level visuo-motor control. This also facilitates the controllable synthesis of accurate, novel sub-goal images. SAGE consists of two key components: (1) a scene graph-based task planner that uses VLMs and LLMs to parse the environment and reason about physically-grounded scene state transition sequences, and (2) a decoupled structural image editing pipeline that controllably converts each target sub-goal graph into a corresponding image through image inpainting and composition. Extensive experiments have demonstrated that SAGE achieves state-of-the-art performance on distinct long-horizon tasks.

Affordance-R1: Reinforcement Learning for Generalizable Affordance Reasoning in Multimodal Large Language Model

Affordance grounding focuses on predicting the specific regions of objects that are associated with the actions to be performed by robots. It plays a vital role in the fields of human-robot interaction, human-object interaction, embodied manipulation, and embodied perception. Existing models often neglect the affordance shared among different objects because they lack the Chain-of-Thought(CoT) reasoning abilities, limiting their out-of-domain (OOD) generalization and explicit reasoning capabilities. To address these challenges, we propose Affordance-R1, the first unified affordance grounding framework that integrates cognitive CoT guided Group Relative Policy Optimization (GRPO) within a reinforcement learning paradigm. Specifically, we designed a sophisticated affordance function, which contains format, perception, and cognition rewards to effectively guide optimization directions. Furthermore, we constructed a high-quality affordance-centric reasoning dataset, ReasonAff, to support training. Trained exclusively via reinforcement learning with GRPO and without explicit reasoning data, Affordance-R1 achieves robust zero-shot generalization and exhibits emergent test-time reasoning capabilities. Comprehensive experiments demonstrate that our model outperforms well-established methods and exhibits open-world generalization. To the best of our knowledge, Affordance-R1 is the first to integrate GRPO-based RL with reasoning into affordance reasoning. The code of our method and our dataset is released on https://github.com/hq-King/Affordance-R1.