Pub Date : 2026-02-11DOI: 10.1126/scirobotics.ady6304
Kevin Holdcroft, Anastasia Bolotnikova, Antoni Jubés Monforte, Jamie Paik
No system is immune to failure. The compromise between reducing failures and improving adaptability is a recurring problem in robotics. Modular robots exemplify this tradeoff, because the number of modules dictates both the possible functions and the odds of failure. We reverse this trend, improving reliability with an increased number of modules by exploiting redundant resources and sharing them locally. We present a unified methodology for local resource sharing; local power sharing balances energy distribution, hybrid communication spreads messages, and local sensor fusion propagates full system state estimate information among the robot collective. We present the experimental results of our methodology applied to a modular robot, Mori3. Despite one module being deprived of its own resources in terms of power, sensing, and communication, the robot collective can successfully perform a locomotion mission in a challenging environment, thanks to neighboring modules supporting each other via our proposed resource-sharing methodology.
{"title":"Scalable robot collective resilience by sharing resources","authors":"Kevin Holdcroft, Anastasia Bolotnikova, Antoni Jubés Monforte, Jamie Paik","doi":"10.1126/scirobotics.ady6304","DOIUrl":"https://doi.org/10.1126/scirobotics.ady6304","url":null,"abstract":"No system is immune to failure. The compromise between reducing failures and improving adaptability is a recurring problem in robotics. Modular robots exemplify this tradeoff, because the number of modules dictates both the possible functions and the odds of failure. We reverse this trend, improving reliability with an increased number of modules by exploiting redundant resources and sharing them locally. We present a unified methodology for local resource sharing; local power sharing balances energy distribution, hybrid communication spreads messages, and local sensor fusion propagates full system state estimate information among the robot collective. We present the experimental results of our methodology applied to a modular robot, Mori3. Despite one module being deprived of its own resources in terms of power, sensing, and communication, the robot collective can successfully perform a locomotion mission in a challenging environment, thanks to neighboring modules supporting each other via our proposed resource-sharing methodology.","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"41 1","pages":""},"PeriodicalIF":25.0,"publicationDate":"2026-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146153726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-11DOI: 10.1126/scirobotics.adx0715
Kun Liang, Rui Wang, Gavin Lyda, Anran Zhang, Wanrong Xie, Yihang Wang, Sicheng Xing, Yizhang Wu, Zhibo Zhang, Yihan Liu, Michael D. Dickey, Bowen Zhu, Wubin Bai
Inspired by the evolutionary diversification of biological eyes for environmental adaptation, recently emerged artificial counterparts offer a variety of visual features that can emulate the eyes of humans, insects, fish, eagles, cats, and others. However, grand challenges reside in developing transformational artificial pupils to address drastic environmental change. Here, we propose a bioinspired vision system that integrates a hemispherical imaging array as an artificial retina with liquid-metal shape-shifters as visual neurons and an adaptive artificial pupil to comprehensively simulate visual recognition with closed-loop pupil reflex behavior. The controlled deformation of the liquid metal allows the design of a range of animal pupil shapes, and the rapid switching of short and open circuits simulates biological spike nerve signals. Under strong light, the system adaptively adjusts the pupil deformation of liquid metal to reduce the amount of exposure, which improves the image recognition accuracy of the artificial vision system under high-light conditions and confirms the key characteristics and functions of the artificial vision system, including ultrawide field of view, adaptive adjustment of light, and image recognition functions. The ability to simulate multiple shapes of animal pupils further demonstrates the programmability of the system and highlights its potential for bioinspired robotic systems, advanced machine vision, and autonomous driving.
{"title":"Bioinspired adaptive pupil reflex based on liquid-metal shape-shifters for machine vision","authors":"Kun Liang, Rui Wang, Gavin Lyda, Anran Zhang, Wanrong Xie, Yihang Wang, Sicheng Xing, Yizhang Wu, Zhibo Zhang, Yihan Liu, Michael D. Dickey, Bowen Zhu, Wubin Bai","doi":"10.1126/scirobotics.adx0715","DOIUrl":"https://doi.org/10.1126/scirobotics.adx0715","url":null,"abstract":"Inspired by the evolutionary diversification of biological eyes for environmental adaptation, recently emerged artificial counterparts offer a variety of visual features that can emulate the eyes of humans, insects, fish, eagles, cats, and others. However, grand challenges reside in developing transformational artificial pupils to address drastic environmental change. Here, we propose a bioinspired vision system that integrates a hemispherical imaging array as an artificial retina with liquid-metal shape-shifters as visual neurons and an adaptive artificial pupil to comprehensively simulate visual recognition with closed-loop pupil reflex behavior. The controlled deformation of the liquid metal allows the design of a range of animal pupil shapes, and the rapid switching of short and open circuits simulates biological spike nerve signals. Under strong light, the system adaptively adjusts the pupil deformation of liquid metal to reduce the amount of exposure, which improves the image recognition accuracy of the artificial vision system under high-light conditions and confirms the key characteristics and functions of the artificial vision system, including ultrawide field of view, adaptive adjustment of light, and image recognition functions. The ability to simulate multiple shapes of animal pupils further demonstrates the programmability of the system and highlights its potential for bioinspired robotic systems, advanced machine vision, and autonomous driving.","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"59 1","pages":""},"PeriodicalIF":25.0,"publicationDate":"2026-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146153663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1126/scirobotics.ady2869
Qi Ye, Qingtao Liu, Siyun Wang, Jiaying Chen, Yu Cui, Ke Jin, Huajin Chen, Xuan Cai, Gaofeng Li, Jiming Chen
Achieving humanlike dexterity with anthropomorphic multifingered robotic hands requires precise finger coordination. However, dexterous manipulation remains highly challenging because of high-dimensional action-observation spaces, complex hand-object contact dynamics, and frequent occlusions. To address this, we drew inspiration from the human learning paradigm of observation and practice and propose a two-stage learning framework by learning visual-tactile integration representations via self-supervised learning from human demonstrations. We trained a unified multitask policy through reinforcement learning and online imitation learning. This decoupled learning enabled the robot to acquire generalizable manipulation skills using only monocular images and simple binary tactile signals. With the unified policy, we built a multifingered hand manipulation system that performs multiple complicated tasks with low-cost sensing. It achieved an 85% success rate across five complex tasks and 25 objects and further generalized to three unseen tasks that share similar hand-object coordination patterns with the training tasks.
{"title":"Visual-tactile pretraining and online multitask learning for humanlike manipulation dexterity","authors":"Qi Ye, Qingtao Liu, Siyun Wang, Jiaying Chen, Yu Cui, Ke Jin, Huajin Chen, Xuan Cai, Gaofeng Li, Jiming Chen","doi":"10.1126/scirobotics.ady2869","DOIUrl":"10.1126/scirobotics.ady2869","url":null,"abstract":"<div >Achieving humanlike dexterity with anthropomorphic multifingered robotic hands requires precise finger coordination. However, dexterous manipulation remains highly challenging because of high-dimensional action-observation spaces, complex hand-object contact dynamics, and frequent occlusions. To address this, we drew inspiration from the human learning paradigm of observation and practice and propose a two-stage learning framework by learning visual-tactile integration representations via self-supervised learning from human demonstrations. We trained a unified multitask policy through reinforcement learning and online imitation learning. This decoupled learning enabled the robot to acquire generalizable manipulation skills using only monocular images and simple binary tactile signals. With the unified policy, we built a multifingered hand manipulation system that performs multiple complicated tasks with low-cost sensing. It achieved an 85% success rate across five complex tasks and 25 objects and further generalized to three unseen tasks that share similar hand-object coordination patterns with the training tasks.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146069929","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1126/scirobotics.adw7868
Xiangxiao Liu, François A. Longchamp, Luca Zunino, Louis Gevers, Lisa R. Schneider, Selina I. Bothner, André Guignard, Alessandro Crespi, Guillaume Bellegarda, Alexandre Bernardino, Eva A. Naumann, Auke J. Ijspeert
Many aquatic animals, including larval zebrafish, exhibit intermittent locomotion, moving via discrete swimming bouts followed by passive glides rather than continuous movement. However, fundamental questions remain unresolved: What neural mechanisms drive this behavior, and what functional benefits does this behavior offer? Specifically, is intermittent swimming more energy efficient than continuous swimming, and, if so, by what mechanism? Live-animal experiments pose technical challenges, because observing or manipulating internal physiological states in freely swimming animals is difficult. Hence, we developed ZBot, a bioinspired robot that replicates the morphological features of larval zebrafish. Embedding a network model inspired by neural circuits and kinematic recordings of larval zebrafish, ZBot reproduces diverse swimming gaits of larval zebrafish bout-and-glide locomotion. By testing ZBot swimming in both turbulent and viscous flow regimes, we confirm that viscous flow markedly reduces traveled distance but minimally affects turning angles. We further tested ZBot in these regimes to analyze how key parameters (tail-beating frequency and amplitude) influence velocity and power use. Our results show that intermittent swimming lowers the energetic cost of transport across most achievable velocities in both flow regimes. Although prior work linked this efficiency to fluid dynamics, like reduced glide drag, we identify an extra mechanism: better actuator efficiency. Mechanistically, this benefit arises because intermittent locomotion shifts the robot’s actuators to higher inherent efficiency. This work introduces a fishlike robot capable of biomimetic intermittent swimming—with demonstrated energy advantages at relevant speeds—and provides general insights into the factors shaping locomotor behavior and efficiency in aquatic animals.
{"title":"Energy efficiency and neural control of continuous versus intermittent swimming in a fishlike robot","authors":"Xiangxiao Liu, François A. Longchamp, Luca Zunino, Louis Gevers, Lisa R. Schneider, Selina I. Bothner, André Guignard, Alessandro Crespi, Guillaume Bellegarda, Alexandre Bernardino, Eva A. Naumann, Auke J. Ijspeert","doi":"10.1126/scirobotics.adw7868","DOIUrl":"10.1126/scirobotics.adw7868","url":null,"abstract":"<div >Many aquatic animals, including larval zebrafish, exhibit intermittent locomotion, moving via discrete swimming bouts followed by passive glides rather than continuous movement. However, fundamental questions remain unresolved: What neural mechanisms drive this behavior, and what functional benefits does this behavior offer? Specifically, is intermittent swimming more energy efficient than continuous swimming, and, if so, by what mechanism? Live-animal experiments pose technical challenges, because observing or manipulating internal physiological states in freely swimming animals is difficult. Hence, we developed ZBot, a bioinspired robot that replicates the morphological features of larval zebrafish. Embedding a network model inspired by neural circuits and kinematic recordings of larval zebrafish, ZBot reproduces diverse swimming gaits of larval zebrafish bout-and-glide locomotion. By testing ZBot swimming in both turbulent and viscous flow regimes, we confirm that viscous flow markedly reduces traveled distance but minimally affects turning angles. We further tested ZBot in these regimes to analyze how key parameters (tail-beating frequency and amplitude) influence velocity and power use. Our results show that intermittent swimming lowers the energetic cost of transport across most achievable velocities in both flow regimes. Although prior work linked this efficiency to fluid dynamics, like reduced glide drag, we identify an extra mechanism: better actuator efficiency. Mechanistically, this benefit arises because intermittent locomotion shifts the robot’s actuators to higher inherent efficiency. This work introduces a fishlike robot capable of biomimetic intermittent swimming—with demonstrated energy advantages at relevant speeds—and provides general insights into the factors shaping locomotor behavior and efficiency in aquatic animals.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146069933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1126/scirobotics.aee5782
Sudharshan Suresh
Visuotactile pretraining with human data leads to robust manipulation policies trained in simulation.
使用人类数据进行视觉预训练,可以在模拟中训练出鲁棒的操作策略。
{"title":"Within arm’s reach: A path forward for robot dexterity","authors":"Sudharshan Suresh","doi":"10.1126/scirobotics.aee5782","DOIUrl":"10.1126/scirobotics.aee5782","url":null,"abstract":"<div >Visuotactile pretraining with human data leads to robust manipulation policies trained in simulation.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146069927","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1126/scirobotics.aee3862
Daniel B. Quinn
The motor efficiency of a zebrafish-like robot helps to explain the advantages of burst-and-coast swimming.
类似斑马鱼的机器人的马达效率有助于解释突发性和海岸游泳的优势。
{"title":"Is intermittent swimming lazy or clever?","authors":"Daniel B. Quinn","doi":"10.1126/scirobotics.aee3862","DOIUrl":"10.1126/scirobotics.aee3862","url":null,"abstract":"<div >The motor efficiency of a zebrafish-like robot helps to explain the advantages of burst-and-coast swimming.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146069928","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-21DOI: 10.1126/scirobotics.aef4218
Amos Matsiko
A soft robotics, self-guided intubation device is capable of fast and safe airway access with minimal user training.
一个软机器人,自我引导插管装置能够快速和安全的气道访问与最少的用户培训。
{"title":"A self-guided intubation device","authors":"Amos Matsiko","doi":"10.1126/scirobotics.aef4218","DOIUrl":"10.1126/scirobotics.aef4218","url":null,"abstract":"<div >A soft robotics, self-guided intubation device is capable of fast and safe airway access with minimal user training.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146020862","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-21DOI: 10.1126/scirobotics.aef4236
Melisa Yashinski
The OriRing achieves a high power-to-weight ratio with origami-inspired joints powered by a soft pneumatic actuator.
OriRing实现了高功率重量比,由软气动执行器驱动的折纸式关节。
{"title":"Lightweight haptic ring delivers high force feedback","authors":"Melisa Yashinski","doi":"10.1126/scirobotics.aef4236","DOIUrl":"10.1126/scirobotics.aef4236","url":null,"abstract":"<div >The OriRing achieves a high power-to-weight ratio with origami-inspired joints powered by a soft pneumatic actuator.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146015370","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Living architectures, such as beehives and ant bridges, adapt continuously to their environments through self-organization of swarming agents. In contrast, most human-made architecture remains static, unable to respond to changing climates or occupant needs. Despite advances in biomimicry within architecture, architectural systems still lack the self-organizing dynamics found in natural swarms. In this work, we introduce the concept of architectural swarms: systems that integrate swarm intelligence and robotics into modular architectural façades to enable responsiveness to environmental conditions and human preferences. We present the Swarm Garden, a proof of concept composed of robotic modules called SGbots. Each SGbot features buckling-sheet actuation, sensing, computation, and wireless communication. SGbots can be networked into reconfigurable spatial systems that exhibit collective behavior, forming a testbed for exploring architectural swarm applications. We demonstrate two application case studies. The first explores adaptive shading using self-organization, where SGbots respond to sunlight using a swarm controller based on opinion dynamics. In a 16-SGbot deployment on an office window, the system adapted effectively to sunlight, showing robustness to sensor failures and different climates. Simulations demonstrated scalability and tunability in larger spaces. The second study explores creative expression in interior design, with 36 SGbots responding to human interaction during a public exhibition, including a live dance performance mediated by a wearable device. Results show that the system was engaging and visually compelling, with 96% positive attendee sentiments. The Swarm Garden exemplifies how architectural swarms can transform the built environment, enabling “living-like” architecture for functional and creative applications.
{"title":"Architectural swarms for responsive façades and creative expression","authors":"Merihan Alhafnawi, Jad Bendarkawi, Yenet Tafesse, Lucia Stein-Montalvo, Azariah Jones, Vicky Chow, Sigrid Adriaenssens, Radhika Nagpal","doi":"10.1126/scirobotics.ady7233","DOIUrl":"10.1126/scirobotics.ady7233","url":null,"abstract":"<div >Living architectures, such as beehives and ant bridges, adapt continuously to their environments through self-organization of swarming agents. In contrast, most human-made architecture remains static, unable to respond to changing climates or occupant needs. Despite advances in biomimicry within architecture, architectural systems still lack the self-organizing dynamics found in natural swarms. In this work, we introduce the concept of architectural swarms: systems that integrate swarm intelligence and robotics into modular architectural façades to enable responsiveness to environmental conditions and human preferences. We present the Swarm Garden, a proof of concept composed of robotic modules called SGbots. Each SGbot features buckling-sheet actuation, sensing, computation, and wireless communication. SGbots can be networked into reconfigurable spatial systems that exhibit collective behavior, forming a testbed for exploring architectural swarm applications. We demonstrate two application case studies. The first explores adaptive shading using self-organization, where SGbots respond to sunlight using a swarm controller based on opinion dynamics. In a 16-SGbot deployment on an office window, the system adapted effectively to sunlight, showing robustness to sensor failures and different climates. Simulations demonstrated scalability and tunability in larger spaces. The second study explores creative expression in interior design, with 36 SGbots responding to human interaction during a public exhibition, including a live dance performance mediated by a wearable device. Results show that the system was engaging and visually compelling, with 96% positive attendee sentiments. The Swarm Garden exemplifies how architectural swarms can transform the built environment, enabling “living-like” architecture for functional and creative applications.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146014884","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1126/scirobotics.adx3017
Yuhang Hu, Jiong Lin, Judah Allen Goldfeder, Philippe M. Wyder, Yifeng Cao, Steven Tian, Yunzhe Wang, Jingran Wang, Mengmeng Wang, Jie Zeng, Cameron Mehlman, Yingke Wang, Delin Zeng, Boyuan Chen, Hod Lipson
Lip motion represents outsized importance in human communication, capturing nearly half of our visual attention during conversation. Yet anthropomorphic robots often fail to achieve lip-audio synchronization, resulting in clumsy and lifeless lip behaviors. Two fundamental barriers underlay this challenge. First, robotic lips typically lack the mechanical complexity required to reproduce nuanced human mouth movements; second, existing synchronization methods depend on manually predefined movements and rules, restricting adaptability and realism. Here, we present a humanoid robot face designed to overcome these limitations, featuring soft silicone lips actuated by a 10–degree-of-freedom mechanism. To achieve lip synchronization without predefined movements, we used a self-supervised learning pipeline based on a variational autoencoder (VAE) combined with a facial action transformer, enabling the robot to autonomously infer more realistic lip trajectories directly from speech audio. Our experimental results suggest that this method outperforms simple heuristics like amplitude-based baselines in achieving more visually coherent lip-audio synchronization. Furthermore, the learned synchronization successfully generalizes across multiple linguistic contexts, enabling robot speech articulation in 10 languages unseen during training.
{"title":"Learning realistic lip motions for humanoid face robots","authors":"Yuhang Hu, Jiong Lin, Judah Allen Goldfeder, Philippe M. Wyder, Yifeng Cao, Steven Tian, Yunzhe Wang, Jingran Wang, Mengmeng Wang, Jie Zeng, Cameron Mehlman, Yingke Wang, Delin Zeng, Boyuan Chen, Hod Lipson","doi":"10.1126/scirobotics.adx3017","DOIUrl":"10.1126/scirobotics.adx3017","url":null,"abstract":"<div >Lip motion represents outsized importance in human communication, capturing nearly half of our visual attention during conversation. Yet anthropomorphic robots often fail to achieve lip-audio synchronization, resulting in clumsy and lifeless lip behaviors. Two fundamental barriers underlay this challenge. First, robotic lips typically lack the mechanical complexity required to reproduce nuanced human mouth movements; second, existing synchronization methods depend on manually predefined movements and rules, restricting adaptability and realism. Here, we present a humanoid robot face designed to overcome these limitations, featuring soft silicone lips actuated by a 10–degree-of-freedom mechanism. To achieve lip synchronization without predefined movements, we used a self-supervised learning pipeline based on a variational autoencoder (VAE) combined with a facial action transformer, enabling the robot to autonomously infer more realistic lip trajectories directly from speech audio. Our experimental results suggest that this method outperforms simple heuristics like amplitude-based baselines in achieving more visually coherent lip-audio synchronization. Furthermore, the learned synchronization successfully generalizes across multiple linguistic contexts, enabling robot speech articulation in 10 languages unseen during training.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 110","pages":""},"PeriodicalIF":27.5,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145964503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}