Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.27865

Frontier LLMs Enter the World of Robotics: A Cross-Model Evaluation on RoboColiseum

Da Huang·Jichao Wang·Junhui Wu·Tong Ran·Linting Guo·Siying Zhang·Chuang Men·Sukai Wang·Yue Zhou·Maoqing Yao·Mo Wu
View original source ↗Read source document ↗Export metadata ↓

Abstract

Recent studies have explored the potential of multimodal large language models for robot control, yet the models' embodied capabilities remain insufficiently characterized across models, tasks, and execution settings. We conduct a multidimensional comparative evaluation on RoboColiseum across four tracks: Instruction Following, Spatial Reasoning, General Manipulation, and Robustness. We implement Agent2Act, a control harness that integrates GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5, and Gemini 4 Argon into the evaluation environment through shared observation, action execution, and feedback interfaces. Under common task settings and scoring criteria, all four models exhibit a consistent capability profile: high instruction-following scores, with three approaching perfect performance; strong spatial reasoning; and general manipulation scores below those of the other three tracks. Within the evaluated configurations, Astra leads instruction following and robustness, while Opus leads spatial reasoning, general manipulation, and overall performance. These results reveal complementary strengths across models and a persistent gap between task understanding and reliable physical execution. Component comparisons on selected tasks further show that harness design affects task success, planning cost, and execution time. Additionally, case studies with Astra examine whole-body coordination, manipulation across robot embodiments, precision manipulation, dexterous-hand manipulation, and real-robot tasks, providing further insight into the model's capabilities and limitations across embodied tasks.

Explore related subjects

cs.RO

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Da Huang, Jichao Wang, Junhui Wu, Tong Ran, Linting Guo, Siying Zhang, Chuang Men, Sukai Wang, Yue Zhou, Maoqing Yao, Mo Wu. 2026-10-20. Frontier LLMs Enter the World of Robotics: A Cross-Model Evaluation on RoboColiseum. https://arxiv.org/abs/2610.27865

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

OEL-AD: An Online Evolutive Learning Framework for Cross-Region Adaptive Autonomous Driving

Autonomous driving models have achieved great progress in recent years, yet most existing approaches remain static, following a train-then-deploy paradigm that lacks the ability to adjust internal parameters when encountering new environments. This limitation is particularly critical in cross-region scenarios, which are common in real-world deployments in autonomous driving where differences in road layouts, traffic patterns, and environmental conditions introduce distribution shifts. In this paper, we argue for a paradigm shift towards Online Evolutive Learning (OEL), a learning paradigm designed to enable adaptive, online model optimization through the interaction and coordination of intelligent components in dynamic environments. We introduce OEL-AD, a novel framework that instantiates this paradigm for autonomous driving. Specifically, OEL-AD leverages trajectory uncertainty from the planning module to determine whether adaptive updates are necessary. When triggered, the framework uses discrepancies between predicted agent behaviors and subsequent observations as a self-supervised learning signal to refine perception and prediction modules, thereby indirectly improving planning accuracy and safety. Experiments on the nuScenes dataset with a cross-region split between Singapore and Boston demonstrate that OEL-AD consistently improves perception and prediction performance. This enhancement in perception and prediction consequently results in improved planning capabilities, thereby confirming the promise of online evolutive learning as a fundamental approach for building adaptive and robust autonomous driving systems.

cs.RO↗
arXiv2026

CENIC: Convex Error-controlled Numerical Integration for Contact

State-of-the-art robotics simulators operate in discrete time. This requires users to choose a time step, which is both critical and challenging: large steps can produce non-physical artifacts, while small steps force the simulation to run slowly. Continuous-time error-controlled integration avoids such issues by automatically adjusting the time step to achieve a desired accuracy. But existing error-controlled integrators struggle with the stiff dynamics of contact, and cannot meet the speed and scalability requirements of modern robotics workflows. We introduce CENIC, a new variable-step integrator that brings together recent advances in convex time-stepping and error-controlled integration, inheriting benefits from both continuous integration and discrete time-stepping. CENIC runs at fast real-time rates comparable to discrete-time robotics simulators like MuJoCo, Drake and Isaac Sim, while also providing guarantees on accuracy and convergence.

cs.RO↗
arXiv2026

Reflection-Based Relative Localization for Cooperative UAV Teams Using Active Markers

Reflections of active markers in the environment are a common source of ambiguity in onboard visual relative localization. This work presents a novel approach that exploits these typically unwanted reflections for onboard relative localization in heterogeneous multi-UAV teams operating over specular, near-planar surfaces such as calm water and flat indoor floors. Using only a camera, active markers, and a single-beam rangefinder, the method requires no prior knowledge of robot size, marker configuration, or surface reflectance properties, and propagates the uncertainty introduced by surface irregularities, including those of dynamic water surfaces. We further characterize, analytically and experimentally, how surface tilt biases the estimate, and identify the slope range over which the level-surface assumption holds. Indoor multi-UAV and outdoor experiments validate the localization accuracy, demonstrating reliable operation across varying lighting conditions and greater effective range than state-of-the-art vision-based active-marker methods, with higher accuracy beyond 30 m. The project website with the video is available under the following link: https://timlakemann.github.io/reflection-based-relative-localization/.

cs.RO↗