المؤلفون: Liu, Ruizhe, Luo, Qian, Yang, Yanchao

مصطلحات موضوعية: Computer Science - Robotics, Computer Science - Artificial Intelligence, Computer Science - Machine Learning

الوصف: We focus on the self-supervised discovery of manipulation concepts that can be adapted and reassembled to address various robotic tasks. We propose that the decision to conceptualize a physical procedure should not depend on how we name it (semantics) but rather on the significance of the informativeness in its representation regarding the low-level physical state and state changes. We model manipulation concepts (discrete symbols) as generative and discriminative goals and derive metrics that can autonomously link them to meaningful sub-trajectories from noisy, unlabeled demonstrations. Specifically, we employ a trainable codebook containing encodings (concepts) capable of synthesizing the end-state of a sub-trajectory given the current state (generative informativeness). Moreover, the encoding corresponding to a particular sub-trajectory should differentiate the state within and outside it and confidently predict the subsequent action based on the gradient of its discriminative score (discriminative informativeness). These metrics, which do not rely on human annotation, can be seamlessly integrated into a VQ-VAE framework, enabling the partitioning of demonstrations into semantically consistent sub-trajectories, fulfilling the purpose of discovering manipulation concepts and the corresponding sub-goal (key) states. We evaluate the effectiveness of the learned concepts by training policies that utilize them as guidance, demonstrating superior performance compared to other baselines. Additionally, our discovered manipulation concepts compare favorably to human-annotated ones while saving much manual effort.
Comment: 27 pages, 15 figures. Published as a conference paper at ICLR 2024

الوصول الحر: http://arxiv.org/abs/2404.10606Test

المؤلفون: Yang, Sizhe, Luo, Qian, Pani, Anumpam, Yang, Yanchao

مصطلحات موضوعية: Computer Science - Robotics

الوصف: Embodied agents capable of complex physical skills can improve productivity, elevate life quality, and reshape human-machine collaboration. We aim at autonomous training of embodied agents for various tasks involving mainly large foundation models. It is believed that these models could act as a brain for embodied agents; however, existing methods heavily rely on humans for task proposal and scene customization, limiting the learning autonomy, training efficiency, and generalization of the learned policies. In contrast, we introduce a brain-body synchronization ({\it BBSEA}) scheme to promote embodied learning in unknown environments without human involvement. The proposed combines the wisdom of foundation models (``brain'') with the physical capabilities of embodied agents (``body''). Specifically, it leverages the ``brain'' to propose learnable physical tasks and success metrics, enabling the ``body'' to automatically acquire various skills by continuously interacting with the scene. We carry out an exploration of the proposed autonomous learning scheme in a table-top setting, and we demonstrate that the proposed synchronization can generate diverse tasks and develop multi-task policies with promising adaptability to new tasks and configurations. We will release our data, code, and trained models to facilitate future studies in building autonomously learning agents with large foundation models in more complex scenarios. More visualizations are available at \href{https://bbsea-embodied-ai.github.ioTest}{https://bbsea-embodied-ai.github.ioTest}

الوصول الحر: http://arxiv.org/abs/2402.08212Test

المؤلفون: Xie, Tianbao, Zhao, Siheng, Wu, Chen Henry, Liu, Yitao, Luo, Qian, Zhong, Victor, Yang, Yanchao, Yu, Tao

مصطلحات موضوعية: Computer Science - Machine Learning, Computer Science - Artificial Intelligence, Computer Science - Computation and Language, Computer Science - Robotics

الوصف: Designing reward functions is a longstanding challenge in reinforcement learning (RL); it requires specialized knowledge or domain data, leading to high costs for development. To address this, we introduce Text2Reward, a data-free framework that automates the generation and shaping of dense reward functions based on large language models (LLMs). Given a goal described in natural language, Text2Reward generates shaped dense reward functions as an executable program grounded in a compact representation of the environment. Unlike inverse RL and recent work that uses LLMs to write sparse reward codes or unshaped dense rewards with a constant function across timesteps, Text2Reward produces interpretable, free-form dense reward codes that cover a wide range of tasks, utilize existing packages, and allow iterative refinement with human feedback. We evaluate Text2Reward on two robotic manipulation benchmarks (ManiSkill2, MetaWorld) and two locomotion environments of MuJoCo. On 13 of the 17 manipulation tasks, policies trained with generated reward codes achieve similar or better task success rates and convergence speed than expert-written reward codes. For locomotion tasks, our method learns six novel locomotion behaviors with a success rate exceeding 94%. Furthermore, we show that the policies trained in the simulator with our method can be deployed in the real world. Finally, Text2Reward further improves the policies by refining their reward functions with human feedback. Video results are available at https://text-to-reward.github.ioTest/ .
Comment: ICLR 2024 camera ready, 37 pages, 12 figures

الوصول الحر: http://arxiv.org/abs/2309.11489Test

المؤلفون: WANG Bing, XIONG Yijing, LUO Qian, ZHONG Wenyi

المصدر: Kouqiang yixue, Vol 44, Iss 6, Pp 475-480 (2024)

مصطلحات موضوعية: children, adolescent, vertical skeletal pattern, sagittal skeletal pattern, temporomandibular joint, Dentistry, RK1-715, Other systems of medicine, RZ201-999

الوصف: Different sagittal and vertical malocclusions exhibit different characteristics in the growth, development, and final morphology of the temporomandibular joint. Different skeletal malocclusions affect the spatial and morphological characteristics of the final temporomandibular joint through different growth and development mechanisms. These mechanisms have important guiding significance for determining the etiology and guiding treatment plans of maxillofacial malocclusion in children and adolescents in clinical practice. This article summarized and analyzed morphological differences of the temporomandibular joint fossa and condyle in different sagittal and vertical malocclusions, as well as the relative position of the condyle in the fossa. It is found that there is a connection between different maxillofacial malocclusions and the characteristics of the temporomandibular joint, with the vertical direction having a more significant impact on the temporomandibular joint than the sagittal direction;the impact of vertical malocclusion on the temporomandibular joint is mainly reflected in the shape of the joint fossa and the position of the condyle in the fossa. The joint fossa of hyperdivergent malocclusion is often relatively low and flat, with the condyle located in the anterior upper position of the fossa. The joint fossa of hypodivergent is relatively narrow and deep, and the condyle is relatively backward and lower in the joint fossa. The possible mechanisms were also elaborated, providing reference for clinicians’ comprehensive diagnosis and treatment.

وصف الملف: electronic resource

العلاقة: https://www.stomatology.cn/fileup/1003-9872/PDF/1719475776887-1219826071.pdfTest; https://doaj.org/toc/1003-9872Test

الوصول الحر: https://doaj.org/article/cc76c8cc3655495db040779a245e7685Test

المؤلفون: Luo, Qian, Li, Yunfei, Wu, Yi

مصطلحات موضوعية: Computer Science - Robotics

الوصف: Grounded understanding of natural language in physical scenes can greatly benefit robots that follow human instructions. In object manipulation scenarios, existing end-to-end models are proficient at understanding semantic concepts, but typically cannot handle complex instructions involving spatial relations among multiple objects. which require both reasoning object-level spatial relations and learning precise pixel-level manipulation affordances. We take an initial step to this challenge with a decoupled two-stage solution. In the first stage, we propose an object-centric semantic-spatial reasoner to select which objects are relevant for the language instructed task. The segmentation of selected objects are then fused as additional input to the affordance learning stage. Simply incorporating the inductive bias of relevant objects to a vision-language affordance learning agent can effectively boost its performance in a custom testbed designed for object manipulation with spatial-related language instructions.
Comment: AAAI 2023 RL Ready for Production Workshop

الوصول الحر: http://arxiv.org/abs/2303.17919Test

المؤلفون: Luo, Qian, Garcia-Menendez, Fernando, Yang, Haozhe, Deshmukh, Ranjit, He, Gang, Lin, Jiang, Johnson, Jeremiah X

المصدر: Environmental Science and Technology. 57(7)

مصطلحات موضوعية: Economics, Applied Economics, Climate Action, Humans, Reproducibility of Results, Air Pollution, Climate, Coal, Power Plants, China, Carbon Dioxide, power system in China, air pollution, public health, Environmental Sciences

الوصف: China's power system is highly regulated and uses an "equal-share" dispatch approach. However, market mechanisms are being introduced to reduce generation costs and improve system reliability. Here, we quantify the climate and human health impacts brought about by this transition, modeling China's power system operations under economic dispatch. We find that significant reductions in mortality related to air pollution (11%) and CO2 emissions (3%) from the power sector can be attained by economic dispatch, relative to the equal-share approach, through more efficient coal-powered generation. Additional health and climate benefits can be achieved by incorporating emission externalities in electricity generation costs. However, the benefits of the transition to economic dispatch will be unevenly distributed across China and may lead to increased health damage in some regions. Our results show the potential of dispatch decision-making in electricity generation to mitigate the negative impacts of power plant emissions with existing facilities in China.

وصف الملف: application/pdf

الوصول الحر: https://escholarship.org/uc/item/2vq7v90qTest
https://escholarship.org/content/qt2vq7v90q/qt2vq7v90q.pdfTest

المؤلفون: HUANG Xin, TENG Lin, NIE Chao-fei, LIU Luo-qian, YAN Feng, QIU Shu-juan, LUO Yu, JIANG Li-long

المصدر: You-qi chuyun, Vol 42, Iss 12, Pp 1337-1351 (2023)

مصطلحات موضوعية: liquid ammonia, methanol, product oil, pipeline, batch transportation, hydrogen energy storage and transportation, Oils, fats, and waxes, TP670-699, Gas industry, TP751-762

الوصف: With the continuous promotion of energy revolution and energy transition, the product oil consumption in China will peak, and the trend of low delivery volume of product oil pipelines becomes evident increasingly, which puts enormous pressure on product oil pipeline companies. At the same time, as the “ammonia-hydrogen” green energy and “liquid sunshine” carbon neutralization technical routes are hotly promoted, and the related industries are developed vigorously, the production of liquid ammonia and methanol is bound to grow massively in the future. Therefore, a technical route for the large-scale batch transportation of liquid ammonia, methanol and other liquid new energy through the existing product oil pipeline systems was proposed, aiming to help the product oil pipeline companies create new growth poles.The batch transportation technology of liquid ammonia/methanol in product oil pipelines was systematically overviewed. Besides, five major technical challenges, including the hydraulic and thermal characteristics in batch transportation, prediction of mixed oil development,batch establishment and mixed oil cutting and treatment, applicability of pipeline materials and equipment, and leakage diffusion and safety protection, were elaborated. This could provide reasonable suggestions and reference directions for the field application of the batch transportation technology of liquid ammonia/methanol/product oil.

وصف الملف: electronic resource

العلاقة: http://kykxxb.cumtb.edu.cn/article/10.6047/j.issn.1000-8241.2023.12.003Test; https://doaj.org/toc/1000-8241Test

الوصول الحر: https://doaj.org/article/a61f9b33ad574955b863ad4cadc61242Test

المؤلفون: Abhyankar, Nikit, Lin, Jiang, Kahrl, Fredrich, Yin, Shengfei, Paliwal, Umed, Liu, Xu, Khanna, Nina, Phadke, Amol A, Luo, Qian

المصدر: iScience. 25(10)

مصطلحات موضوعية: Environmental Sciences, Environmental Management, Affordable and Clean Energy, Energy Modelling, Energy management, Energy policy, Energy resources, Energy sustainability

الوصف: Dramatic reductions in solar, wind, and battery storage costs create new opportunities to reduce emissions and costs in China’s electricity sector, beyond current policy goals. This study examines the cost, reliability, emissions, public health, and employment implications of increasing the share of non-fossil fuel (“carbon free”) electricity generation in China to 80% by 2035. The analysis uses state-of-the-art modeling with high resolution load, wind, and solar inputs. The study finds that achieving an 80% carbon free electricity system in China by 2035 could reduce wholesale electricity costs, relative to a current policy baseline, while maintaining high levels of reliability, reducing deaths from air pollution, and increasing employment. In our 80% scenario, wind and solar generation capacity reach 3 TW and battery storage capacity reaches 0.4 TW by 2035, implying a rapid scale up in these resources that will require changes in policy targets, markets and regulation, and land use policies.

وصف الملف: application/pdf

الوصول الحر: https://escholarship.org/uc/item/9183b502Test
https://escholarship.org/content/qt9183b502/qt9183b502.pdfTest

المؤلفون: XING Zhiwei, KAN Ben, LIU Zishuo, LI Biao, LUO Qian

المصدر: Shanghai Jiaotong Daxue xuebao, Vol 57, Iss 10, Pp 1292-1304 (2023)

مصطلحات موضوعية: pavement snow and ice state perception, yolox-s, global context block (gc block), bi-directional feature pyramid network (bifpn), adaptive spatial feature fusion (asff), α-eiou loss function, Engineering (General). Civil engineering (General), TA1-2040, Chemical engineering, TP155-156, Naval architecture. Shipbuilding. Marine engineering, VM1-989

الوصف: Aimed at the lack of awarness of safety and airworthiness state perception ability of airport ice runway and the new demand of interaction of runway surface condition report, a multi-scale feature fusion based ice and snow state perception model of airport runway is proposed. Based on the YOLOX-s model, first, the global context block (GC block) is introduced into the backbone feature extraction network to obtain more abundant shallow and deep features. Then, the PANet networks in neck are replaced with the bi-directional feature pyramid network (BiFPN) to improve the feature fusion ability. Afterwards, an adaptive spatial feature fusion (ASFF) structure is added to the tail of the enhanced feature extraction network to further enhance the feature fusion effect. Finally, α-EIoU is used to optimize the loss function to improve the convergence speed and accuracy of the model. The experimental results show that the improved YOLOX-s model has an average accuracy of 91.53% in the snow and ice pollutant data set obtained from the runway snow and ice experimental system, which is 4.68% higher than the original YOLOX-s model, and can provide decision-making support for airport runway snow removal operations.

وصف الملف: electronic resource

العلاقة: https://xuebao.sjtu.edu.cn/article/2023/1006-2467/1006-2467-57-10-1292.shtmlTest; https://doaj.org/toc/1006-2467Test

الوصول الحر: https://doaj.org/article/b3bff95ec48d43169db104c3d28be5e6Test

المؤلفون: Yokoyama, Naoki, Luo, Qian, Batra, Dhruv, Ha, Sehoon

مصطلحات موضوعية: Computer Science - Robotics, Computer Science - Artificial Intelligence

الوصف: Recent advances in deep reinforcement learning and scalable photorealistic simulation have led to increasingly mature embodied AI for various visual tasks, including navigation. However, while impressive progress has been made for teaching embodied agents to navigate static environments, much less progress has been made on more dynamic environments that may include moving pedestrians or movable obstacles. In this study, we aim to benchmark different augmentation techniques for improving the agent's performance in these challenging environments. We show that adding several dynamic obstacles into the scene during training confers significant improvements in test-time generalization, achieving much higher success rates than baseline agents. We find that this approach can also be combined with image augmentation methods to achieve even higher success rates. Additionally, we show that this approach is also more robust to sim-to-sim transfer than image augmentation methods. Finally, we demonstrate the effectiveness of this dynamic obstacle augmentation approach by using it to train an agent for the 2021 iGibson Challenge at CVPR, where it achieved 1st place for Interactive Navigation. Video link: https://www.youtube.com/watch?v=HxUX2HeOSE4Test

الوصول الحر: http://arxiv.org/abs/2109.10493Test