Robotic manipulation policies often fail to generalize because they must simultaneously learn where to attend, what actions to take, and how to execute them. We argue that high-level reasoning about where and what can be offloaded to vision-language models (VLMs), leaving policies to specialize in how to act. We present PEEK (Policy-agnostic Extraction of Essential Keypoints), which fine-tunes VLMs to predict a unified point-based intermediate representation: 1. end-effector paths specifying what actions to take, and 2. task-relevant masks indicating where to focus. These annotations are directly overlaid onto robot observations, making the representation policy-agnostic and transferable across architectures. To enable scalable training, we introduce an automatic annotation pipeline, generating labeled data across 20+ robot datasets spanning 9 embodiments. In real-world evaluations, PEEK consistently boosts zero-shot generalization, including a 41.4x real-world improvement for a 3D policy trained only in simulation, and 2-3.5x gains for both large VLAs and small manipulation policies. By letting VLMs absorb semantic and visual complexity, PEEK equips manipulation policies with the minimal cues they need--where, what, and how. Website at https://peek-robot.github.io/.

Large multimodal models (LMMs) encode universal physical laws observed during training, such as momentum conservation, as parametric knowledge. It allows LMMs to answer physical reasoning queries, such as the outcome of a potential collision event from visual input. However, since parametric knowledge includes only the physical laws seen during training, it is insufficient for reasoning when the inference scenario violates these physical laws. In contrast, humans possess the skill to adapt their physical reasoning to unseen physical environments from a few visual examples. This ability, which we refer to as inductive physical reasoning, is indispensable for LMMs if they are to replace human agents in safety-critical applications. Despite its importance, existing visual benchmarks evaluate only the parametric knowledge in LMMs, and not inductive physical reasoning. To this end, we propose InPhyRe, the first visual question answering benchmark to measure inductive physical reasoning in LMMs. InPhyRe evaluates LMMs on their ability to predict the outcome of collision events in algorithmically generated synthetic collision videos. By inspecting 13 LMMs, InPhyRe informs us that (1) LMMs struggle to apply their limited parametric knowledge about universal physical laws to reasoning, (2) inductive physical reasoning in LMMs is weak when demonstration samples violate universal physical laws, and (3) inductive physical reasoning in LMMs suffers from language bias and largely ignores the visual inputs, questioning the trustworthiness of LMMs regarding visual inputs.

Recent advancements in Large Language Models (LLMs) and Visual Language Models (VLMs) have significantly impacted robotics, enabling high-level semantic motion planning applications. Reinforcement Learning (RL), a complementary paradigm, enables agents to autonomously optimize complex behaviors through interaction and reward signals. However, designing effective reward functions for RL remains challenging, especially in real-world tasks where sparse rewards are insufficient and dense rewards require elaborate design. In this work, we propose Autonomous Reinforcement learning for Complex HumanInformed Environments (ARCHIE), an unsupervised pipeline leveraging GPT-4, a pre-trained LLM, to generate reward functions directly from natural language task descriptions. The rewards are used to train RL agents in simulated environments, where we formalize the reward generation process to enhance feasibility. Additionally, GPT-4 automates the coding of task success criteria, creating a fully automated, one-shot procedure for translating human-readable text into deployable robot skills. Our approach is validated through extensive simulated experiments on single-arm and bi-manual manipulation tasks using an ABB YuMi collaborative robot, highlighting its practicality and effectiveness. Tasks are demonstrated on the real robot setup.

Traditional reinforcement learning-based robotic control methods are often task-specific and fail to generalize across diverse environments or unseen objects and instructions. Visual Language Models (VLMs) demonstrate strong scene understanding and planning capabilities but lack the ability to generate actionable policies tailored to specific robotic embodiments. To address this, Visual-Language-Action (VLA) models have emerged, yet they face challenges in long-horizon spatial reasoning and grounded task planning. In this work, we propose the Embodied Multimodal Action Model with Grounded Chain of Thought and Look-ahead Spatial Reasoning, Emma-X. Emma-X leverages our constructed hierarchical embodiment dataset based on BridgeV2, containing 60,000 robot manipulation trajectories auto-annotated with grounded task reasoning and spatial guidance. Additionally, we introduce a trajectory segmentation strategy based on gripper states and motion trajectories, which can help mitigate hallucination in grounding subtask reasoning generation. Experimental results demonstrate that Emma-X achieves superior performance over competitive baselines, particularly in real-world robotic tasks requiring spatial reasoning.

Robotic systems for manipulation tasks are increasingly expected to be easy to configure for new tasks or unpredictable environments, while keeping a transparent policy that is readable and verifiable by humans. We propose the method BEhavior TRee eXPansion with Large Language Models (BETR-XP-LLM) to dynamically and automatically expand and configure Behavior Trees as policies for robot control. The method utilizes an LLM to resolve errors outside the task planner's capabilities, both during planning and execution. We show that the method is able to solve a variety of tasks and failures and permanently update the policy to handle similar problems in the future.

Current robot autonomy struggles to operate beyond the assumed Operational Design Domain (ODD), the specific set of conditions and environments in which the system is designed to function, while the real-world is rife with uncertainties that may lead to failures. Automating recovery remains a significant challenge. Traditional methods often rely on human intervention to manually address failures or require exhaustive enumeration of failure cases and the design of specific recovery policies for each scenario, both of which are labor-intensive. Foundational Vision-Language Models (VLMs), which demonstrate remarkable common-sense generalization and reasoning capabilities, have broader, potentially unbounded ODDs. However, limitations in spatial reasoning continue to be a common challenge for many VLMs when applied to robot control and motion-level error recovery. In this paper, we investigate how optimizing visual and text prompts can enhance the spatial reasoning of VLMs, enabling them to function effectively as black-box controllers for both motion-level position correction and task-level recovery from unknown failures. Specifically, the optimizations include identifying key visual elements in visual prompts, highlighting these elements in text prompts for querying, and decomposing the reasoning process for failure detection and control generation. In experiments, prompt optimizations significantly outperform pre-trained Vision-Language-Action Models in correcting motion-level position errors and improve accuracy by 65.78% compared to VLMs with unoptimized prompts. Additionally, for task-level failures, optimized prompts enhanced the success rate by 5.8%, 5.8%, and 7.5% in VLMs' abilities to detect failures, analyze issues, and generate recovery plans, respectively, across a wide range of unknown errors in Lego assembly.

Generative models trained on internet data have revolutionized how text, image, and video content can be created. Perhaps the next milestone for generative models is to simulate realistic experience in response to actions taken by humans, robots, and other interactive agents. Applications of a real-world simulator range from controllable content creation in games and movies, to training embodied agents purely in simulation that can be directly deployed in the real world. We explore the possibility of learning a universal simulator of real-world interaction through generative modeling. We first make the important observation that natural datasets available for learning a real-world simulator are often rich along different dimensions (e.g., abundant objects in image data, densely sampled actions in robotics data, and diverse movements in navigation data). With careful orchestration of diverse datasets, each providing a different aspect of the overall experience, we can simulate the visual outcome of both high-level instructions such as "open the drawer" and low-level controls such as "move by x, y" from otherwise static scenes and objects. We use the simulator to train both high-level vision-language policies and low-level reinforcement learning policies, each of which can be deployed in the real world in zero shot after training purely in simulation. We also show that other types of intelligence such as video captioning models can benefit from training with simulated experience, opening up even wider applications. Video demos can be found at universal-simulator.github.io.

Explaining the behaviour of intelligent agents learned by reinforcement learning (RL) to humans is challenging yet crucial due to their incomprehensible proprioceptive states, variational intermediate goals, and resultant unpredictability. Moreover, one-step explanations for RL agents can be ambiguous as they fail to account for the agent's future behaviour at each transition, adding to the complexity of explaining robot actions. By leveraging abstracted actions that map to task-specific primitives, we avoid explanations on the movement level. To further improve the transparency and explainability of robotic systems, we propose an explainable Q-Map learning framework that combines reward decomposition (RD) with abstracted action spaces, allowing for non-ambiguous and high-level explanations based on object properties in the task. We demonstrate the effectiveness of our framework through quantitative and qualitative analysis of two robotic scenarios, showcasing visual and textual explanations, from output artefacts of RD explanations, that are easy for humans to comprehend. Additionally, we demonstrate the versatility of integrating these artefacts with large language models (LLMs) for reasoning and interactive querying.

We propose a novel framework for learning high-level cognitive capabilities in robot manipulation tasks, such as making a smiley face using building blocks. These tasks often involve complex multi-step reasoning, presenting significant challenges due to the limited paired data connecting human instructions (e.g., making a smiley face) and robot actions (e.g., end-effector movement). Existing approaches relieve this challenge by adopting an open-loop paradigm decomposing high-level instructions into simple sub-task plans, and executing them step-by-step using low-level control models. However, these approaches are short of instant observations in multi-step reasoning, leading to sub-optimal results. To address this issue, we propose to automatically collect a cognitive robot dataset by Large Language Models (LLMs). The resulting dataset AlphaBlock consists of 35 comprehensive high-level tasks of multi-step text plans and paired observation sequences. To enable efficient data acquisition, we employ elaborated multi-round prompt designs that effectively reduce the burden of extensive human involvement. We further propose a closed-loop multi-modal embodied planning model that autoregressively generates plans by taking image observations as input. To facilitate effective learning, we leverage MiniGPT-4 with a frozen visual encoder and LLM, and finetune additional vision adapter and Q-former to enable fine-grained spatial perception for manipulation tasks. We conduct experiments to verify the superiority over existing open and closed-loop methods, and achieve a significant increase in success rate by 21.4% and 14.5% over ChatGPT and GPT-4 based robot tasks. Real-world demos are shown in https://www.youtube.com/watch?v=ayAzID1_qQk .

DALL·E 2 is the newest AI model by OpenAI. If you've seen some of its creations and think they're amazing, keep reading to understand why you're totally right -- but also wrong. OpenAI published a blog post and a paper entitled "Hierarchical Text-Conditional Image Generation with CLIP Latents" on DALL·E 2. The post is fine if you want to get a glimpse at the results and the paper is great for understanding the technical details, but neither explains DALL·E 2's amazingness -- and the not-so-amazing -- in depth. That's what this article is for. If this in-depth educational content is useful for you, subscribe to our AI mailing list to be alerted when we release new material. DALL·E 2 is the new version of DALL·E, a generative language model that takes sentences and creates corresponding original images. At 3.5B parameters, DALL·E 2 is a large model but not nearly as large as GPT-3 and, interestingly, smaller than its predecessor (12B). Despite its size, DALL·E 2 generates 4x better resolution images than DALL·E and it's preferred by human judges 70% of the time both in caption matching and photorealism. As they did with DALL·E, OpenAI didn't release DALL·E 2 (you can always join the never-ending waitlist). However, they open-sourced CLIP which, although only indirectly related to DALL·E, forms the basis of DALL·E 2. (CLIP is also the basis of the apps and notebooks people who can't access DALL·E 2 are using.)