Chemistry laboratory automation aims to increase throughput, reproducibility, and safety, yet many existing systems still depend on frequent human intervention. Advances in robotics have reduced this dependency, but without a structured representation of the required skills, autonomy remains limited to bespoke, task-specific solutions with little capacity to transfer beyond their initial design. Current experiment abstractions typically describe protocol-level steps without specifying the robotic actions needed to execute them. This highlights the lack of a systematic account of the manipulation skills required for robots in chemistry laboratories. To address this gap, we introduce TARMAC - a Taxonomy for Robot Manipulation in Chemistry - a domain-specific framework that defines and organizes the core manipulations needed in laboratory practice. Based on annotated teaching-lab demonstrations and supported by experimental validation, TARMAC categorizes actions according to their functional role and physical execution requirements. Beyond serving as a descriptive vocabulary, TARMAC can be instantiated as robot-executable primitives and composed into higher-level macros, enabling skill reuse and supporting scalable integration into long-horizon workflows. These contributions provide a structured foundation for more flexible and autonomous laboratory automation. More information is available at https://tarmac-paper.github.io/

Robotic chemists promise to both liberate human experts from repetitive tasks and accelerate scientific discovery, yet remain in their infancy. Chemical experiments involve long-horizon procedures over hazardous and deformable substances, where success requires not only task completion but also strict compliance with experimental norms. To address these challenges, we propose \textit{RoboChemist}, a dual-loop framework that integrates Vision-Language Models (VLMs) with Vision-Language-Action (VLA) models. Unlike prior VLM-based systems (e.g., VoxPoser, ReKep) that rely on depth perception and struggle with transparent labware, and existing VLA systems (e.g., RDT, pi0) that lack semantic-level feedback for complex tasks, our method leverages a VLM to serve as (1) a planner to decompose tasks into primitive actions, (2) a visual prompt generator to guide VLA models, and (3) a monitor to assess task success and regulatory compliance. Notably, we introduce a VLA interface that accepts image-based visual targets from the VLM, enabling precise, goal-conditioned control. Our system successfully executes both primitive actions and complete multi-step chemistry protocols. Results show 23.57% higher average success rate and a 0.298 average increase in compliance rate over state-of-the-art VLA baselines, while also demonstrating strong generalization to objects and tasks.

Chemistry experimentation is often resource- and labor-intensive. Despite the many benefits incurred by the integration of advanced and special-purpose lab equipment, many aspects of experimentation are still manually conducted by chemists, for example, polishing an electrode in electrochemistry experiments. Traditional lab automation infrastructure faces challenges when it comes to flexibly adapting to new chemistry experiments. To address this issue, we propose a human-friendly and flexible robotic system, ORGANA, that automates a diverse set of chemistry experiments. It is capable of interacting with chemists in the lab through natural language, using Large Language Models (LLMs). ORGANA keeps scientists informed by providing timely reports that incorporate statistical analyses. Additionally, it actively engages with users when necessary for disambiguation or troubleshooting. ORGANA can reason over user input to derive experiment goals, and plan long sequences of both high-level tasks and low-level robot actions while using feedback from the visual perception of the environment. It also supports scheduling and parallel execution for experiments that require resource allocation and coordination between multiple robots and experiment stations. We show that ORGANA successfully conducts a diverse set of chemistry experiments, including solubility assessment, pH measurement, recrystallization, and electrochemistry experiments. For the latter, we show that ORGANA robustly executes a long-horizon plan, comprising 19 steps executed in parallel, to characterize the electrochemical properties of quinone derivatives, a class of molecules used in rechargeable flow batteries. Our user study indicates that ORGANA significantly improves many aspects of user experience while reducing their physical workload. More details about ORGANA can be found at https://ac-rad.github.io/organa/.

Large language models offer new ways of empowering people to program robot applications-namely, code generation via prompting. However, the code generated by LLMs is susceptible to errors. This work reports a preliminary exploration that empirically characterizes common errors produced by LLMs in robot programming. We categorize these errors into two phases: interpretation and execution. In this work, we focus on errors in execution and observe that they are caused by LLMs being "forgetful" of key information provided in user prompts. Based on this observation, we propose prompt engineering tactics designed to reduce errors in execution. We then demonstrate the effectiveness of these tactics with three language models: ChatGPT, Bard, and LLaMA-2. Finally, we discuss lessons learned from using LLMs in robot programming and call for the benchmarking of LLM-powered end-user development of robot applications.

Chain-of-thought (CoT) prompting, which offers step-by-step problem-solving rationales, has impressively unlocked the reasoning potential of large language models (LLMs). Yet, the standard CoT is less effective in problems demanding multiple reasoning steps. This limitation arises from the complex reasoning process in multi-step problems: later stages often depend on the results of several steps earlier, not just the results of the immediately preceding step. Such complexities suggest the reasoning process is naturally represented as a graph. The almost linear and straightforward structure of CoT prompting, however, struggles to capture this complex reasoning graph. To address this challenge, we propose Residual Connection Prompting (RESPROMPT), a new prompting strategy that advances multi-step reasoning in LLMs. Our key idea is to reconstruct the reasoning graph within prompts. We achieve this by integrating necessary connections-links present in the reasoning graph but missing in the linear CoT flow-into the prompts. Termed "residual connections", these links are pivotal in morphing the linear CoT structure into a graph representation, effectively capturing the complex reasoning graphs inherent in multi-step problems. We evaluate RESPROMPT on six benchmarks across three diverse domains: math, sequential, and commonsense reasoning. For the open-sourced LLaMA family of models, RESPROMPT yields a significant average reasoning accuracy improvement of 12.5% on LLaMA-65B and 6.8% on LLaMA2-70B. Breakdown analysis further highlights RESPROMPT particularly excels in complex multi-step reasoning: for questions demanding at least five reasoning steps, RESPROMPT outperforms the best CoT based benchmarks by a remarkable average improvement of 21.1% on LLaMA-65B and 14.3% on LLaMA2-70B. Through extensive ablation studies and analyses, we pinpoint how to most effectively build residual connections.

Keeping track of how states of entities change as a text or dialog unfolds is a key prerequisite to discourse understanding. Yet, there have been few systematic investigations into the ability of large language models (LLMs) to track discourse entities. In this work, we present a task probing to what extent a language model can infer the final state of an entity given an English description of the initial state and a series of state-changing operations. We use this task to first investigate whether Flan-T5, GPT-3 and GPT-3.5 can track the state of entities, and find that only GPT-3.5 models, which have been pretrained on large amounts of code, exhibit this ability. We then investigate whether smaller models pretrained primarily on text can learn to track entities, through finetuning T5 on several training/evaluation splits. While performance degrades for more complex splits, we find that even when evaluated on a different set of entities from training or longer operation sequences, a finetuned model can perform non-trivial entity tracking. Taken together, these results suggest that language models can learn to track entities but pretraining on text corpora alone does not make this capacity surface.

Chemists need to perform many laborious and time-consuming experiments in the lab to discover and understand the properties of new materials. To support and accelerate this process, we propose a robot framework for manipulation that autonomously performs chemistry experiments. Our framework receives high-level abstract descriptions of chemistry experiments, perceives the lab workspace, and autonomously plans multi-step actions and motions. The robot interacts with a wide range of lab equipment and executes the generated plans. A key component of our method is constrained task and motion planning using PDDLStream solvers. Preventing collisions and spillage is done by introducing a constrained motion planner. Our planning framework can conduct different experiments employing implemented actions and lab tools. We demonstrate the utility of our framework on pouring skills for various materials and two fundamental chemical experiments for materials synthesis: solubility and recrystallization.

Generating low-level robot task plans from high-level natural language instructions remains a challenging problem. Although large language models have shown promising results in generating plans, the accuracy of the output remains unverified. Furthermore, the lack of domain-specific language data poses a limitation on the applicability of these models. In this paper, we propose CLAIRIFY, a novel approach that combines automatic iterative prompting with program verification to ensure programs written in data-scarce domain-specific language are syntactically valid and incorporate environment constraints. Our approach provides effective guidance to the language model on generating structured-like task plans by incorporating any errors as feedback, while the verifier ensures the syntactic accuracy of the generated plans. We demonstrate the effectiveness of CLAIRIFY in planning chemistry experiments by achieving state-of-the-art results. We also show that the generated plans can be executed on a real robot by integrating them with a task and motion planner.

More than one hundred benchmarks have been developed to test the commonsense knowledge and commonsense reasoning abilities of artificial intelligence (AI) systems. However, these benchmarks are often flawed and many aspects of common sense remain untested. Consequently, we do not currently have any reliable way of measuring to what extent existing AI systems have achieved these abilities. This paper surveys the development and uses of AI commonsense benchmarks. We discuss the nature of common sense; the role of common sense in AI; the goals served by constructing commonsense benchmarks; and desirable features of commonsense benchmarks. We analyze the common flaws in benchmarks, and we argue that it is worthwhile to invest the work needed ensure that benchmark examples are consistently high quality. We survey the various methods of constructing commonsense benchmarks. We enumerate 139 commonsense benchmarks that have been developed: 102 text-based, 18 image-based, 12 video based, and 7 simulated physical environments. We discuss the gaps in the existing benchmarks and aspects of commonsense reasoning that are not addressed in any existing benchmark. We conclude with a number of recommendations for future development of commonsense AI benchmarks.

Unlike most existing question-answering (QA) datasets, we expect models to not only answer questions, but also produce step-by-step structured explanations describing how premises in the question are used to produce intermediate conclusions that can prove the correctness of a certain answer. We perform extensive evaluation with popular language models such as few-shot prompting GPT-3 and fine-tuned T5. We find that these models still lag behind human performance when producing such structured reasoning steps. We believe this work will provide a way for the community to better train and test systems on multi-step reasoning and explanations in natural language. A long-term pursuit in Artificial Intelligence is to endow machines with the ability to reason and manipulate premises to reach conclusions and perform tasks. Some recent works in the field of question-answering (QA) have demonstrated that language models can bypass some of these issues and learn to reason directly over natural language (Clark et al., 2020), allowing for more flexible and adaptable reasoning capabilities. Another advantage of performing multi-step reasoning over natural language is that it allows for more inspectable outputs, improving the explainability of models that are otherwise regarded as black box systems (Jain & Wallace, 2019; Rajani et al., 2019a; Danilevsky et al., 2020). Despite the recent progress, we notice that there is still a gap in resources for training and evaluating general reasoning capabilities over natural language. We build upon existing QA datasets by adding multi-premise, multi-step, structured explanations in the form of reasoning graphs, as depicted in Figure 1. When combined, all reasoning graphs contain a total of 151.1k reasoning steps (or textual entailments), of which 14.7k were created by our expert annotators.