USC at ICRA 2026 (Credit: Dall-E)
USC researchers will present 32 papers at the 2026 IEEE International Conference on Robotics and Automation (ICRA), with research spanning dexterous manipulation, safe autonomous navigation, vision-language models for robot learning, robots in space and bio-inspired robots.
Held June 1–5 in Vienna, Austria, ICRA is one of the world’s leading and largest conferences in robotics and automation, bringing together researchers, engineers and industry leaders to share advances in artificial intelligence (AI), autonomous systems, human-robot interaction and emerging robotics technologies.
As a growing leader in robotics research and AI innovation, USC has faculty and students from labs across the university attending, including researchers from the USC Viterbi School of Engineering’s Thomas Lord Department of Computer Science (CS), the Ming Hsieh Department of Electrical and Computer Engineering (ECE) and the Department of Astronautical Engineering, and USC Mark and Mary Stevens School of Computing and Artificial Intelligence.
USC researchers will present papers across poster, workshop and highly selective oral sessions, as well as lead and deliver talks at various workshops. Yue Wang, assistant professor of computer science, will give an invited talk at the “Beyond Teleoperation: Learning from Diverse Human and Simulation Data” workshop on the final day of the conference. Many faculty members and labs also have multiple papers accepted, including Daniel Seita’s Sensing, Learning and Understanding for Robotic Manipulation (SLURM) Lab, which had seven papers accepted, several of them in collaboration with other USC Viterbi researchers like Gaurav Sukhatme, executive vice dean, director of the USC Mark and Mary Stevens School of Computing and AI and incoming interim dean of USC Viterbi’s School of Engineering; Seita is an assistant professor of computer science.
Keenan Albee, assistant professor of astronautical engineering and aerospace and mechanical engineering, is an organizer and program chair for the “Intelligent Space Robotics and Systems” workshop held on the final day of the conference. Ishika Singh, a PhD student advised by Jesse Thomason, is also one of the organizers for the conference’s VLA Pipeline session. Thomason is an assistant professor of computer science.
Erdem Bıyık will also be delivering a talk at the workshop, “Bridging the Gap between Robot Learning and Human-Robot Interaction,” as an invited speaker on the last day of the conference. Bıyık is an assistant professor of computer science and electrical and computer engineering with joint appointments at
Beginning today, the conference continues to grow increasingly competitive, with the paper acceptance rate dropping to 35% this year from 38% last year. More than 5,000 papers were submitted, with only 1,800 accepted in 2026.
Organized annually by the Institute of Electrical and Electronics Engineers and its Robotics and Automation Society, the conference features peer-reviewed research papers, workshops, tutorials and technical demonstrations from institutions around the world.
USC @ ICRA 2026 Research Spotlights
Learning From Lizards: Meet Tail-Powered Robots for Sandy Terrain
USC researchers are looking to the animal kingdom to solve one of the most frustrating problems in robotics: how to keep a robot from getting stuck in the sand. In the paper “Bio-Inspired Tail Oscillation Enables Fast Crawling on Deformable Granular Terrains,” the team demonstrates that adding a swinging, oscillating tail to a crawling robot significantly boosts its speed and stability.
Much like lizards and other tail-bearing animals, the robot uses the rhythmic movement of its tail to maintain momentum and balance on deformable surfaces such as sand or gravel that would otherwise cause robots to slip or sink. The breakthrough suggests that for robots designed for desert exploration or search-and-rescue missions in unstable environments, the secret to navigating difficult terrain may lie in a bit of tail-wagging.
Safer Autonomous Vehicles on the Road and in the Air: Lowering Collisions, Enabling Multi-Robot Collaboration and Navigating Unpredictable Spaces
USC engineers are developing methods that help autonomous systems navigate complex, unknown and dynamic environments without collisions, spanning aerial drones, quadrupeds and bipedal robots. Several papers focus on improving how robots plan, adapt and coordinate movement under uncertainty, with an emphasis on safety in real-world conditions.
In “Latent Activation Editing: Inference-Time Refinement of Learned Policies for Safer Multirobot Navigation,” researchers introduce a framework that improves the safety of pre-trained navigation policies without modifying the underlying model. By monitoring internal activations in real time, the system identifies risky states and adjusts behavior toward safer trajectories during inference. In multi-quadrotor experiments, the method reduced cumulative collisions by nearly 90% while preserving successful task completion.
Other work focuses on enabling robust navigation for legged systems operating in unpredictable terrain. “One Filter to Deploy Them All: Robust Safety for Quadrupedal Navigation in Unknown Environments” develops a safety framework for quadruped robots moving through unmapped spaces, while “STATE-NAV: Stability-Aware Traversability Estimation for Bipedal Navigation on Rough Terrain” helps bipedal robots assess which paths are safe to traverse on uneven or unstable ground.
Additionally, studies advance formal safety and coordination in autonomous systems. “Safety Evaluation of Motion Plans Using Trajectory Predictors As Forward Reachable Set Estimators” introduces a predictive approach for verifying whether a robot’s planned trajectory could lead to collisions by estimating all possible future states. Meanwhile, “UMBRELLA: Uncertainty-Aware Multi-Robot Reactive Coordination Under Dynamic Temporal Logic Tasks” enables multiple robots to coordinate safely under uncertainty and evolving task constraints. Rounding out the set, “DualGuard MPPI: Safe and Performant Optimal Control by Combining Sampling-Based MPC and Hamilton-Jacobi Reachability” integrates two control paradigms to balance high performance with mathematically grounded safety guarantees, and “Learning to Drive Anywhere With Model-Based Reannotation” extends generalizable navigation to previously unseen environments, pushing forward the goal of truly robust autonomous mobility.
From Factory Floors to Everyday Tasks: Teaching Robots to Work With Their Hands Like Humans
From assembling industrial components to scooping liquids and following instruction manuals, USC researchers are advancing robotic systems that can interact with the physical world with greater precision, adaptability and human-like dexterity. Many papers explore how robots can better manipulate objects in complex, unpredictable environments– a longstanding challenge in robotics.
In the paper “Refinery: Active Fine-Tuning and Deployment-Time Optimization for Contact-Rich Policies,” researchers tackle one of the biggest limitations in robotic assembly: reliability. While many simulation-trained robots can complete assembly tasks with roughly 80% success rates, that level of accuracy remains too fragile for industrial settings where robots must complete multiple tasks in sequence. The team developed a framework called Refinery, which improves robotic assembly success rates to more than 91% in simulation and enables robots to chain together the assembly of up to eight parts without explicit multi-step training.
Other USC projects focus on helping robots perform more nuanced and flexible forms of manipulation. In “Learning Geometry-Aware Nonprehensile Pushing and Pulling with Dexterous Hands,” researchers study how robots can move objects through pushing and pulling motions without fully grasping them, allowing for more human-like interactions with objects. Meanwhile, “SCOOP’D: State-based Sim2Real Generative Policy for Generalizable Mixed-Liquid-Solid Scooping” explores how robots can manipulate challenging materials that combine liquids and solids, improving how robotic systems transfer skills learned in simulation into real-world environments.
Several projects also investigate how robots can learn directly from humans and adapt to open-world settings. “Manual2Skill++: Connector-Aware General Robotic Assembly from Instruction Manuals Via Vision-Language Models” enables robots to interpret instruction manuals using vision-language models to complete complex assembly tasks. Other work, including “OCRA: Object-Centric Learning with 3D and Tactile Priors for Human-to-Robot Action Transfer” and “Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation,” focuses on tactile learning, human-to-robot action transfer and large-scale datasets designed to help humanoid robots navigate everyday manipulation tasks in real-world environments.
Teaching Robots to Think, See and Plan With Vision-Language and AI Learning
USC researchers are advancing how robots interpret the world and make decisions by combining large language models (LLMs), vision-language models (VLMs) and new forms of machine learning that bridge perception and action. Several papers presented at ICRA 2026 explore how these AI systems can help robots better understand complex environments, plan more effectively and adapt to new tasks with less human input.
In “IMPACT: Intelligent Motion Planning with Acceptable Contact Trajectories via Vision-Language Models,” researchers use VLMs to help robots plan motion trajectories that intentionally include controlled, acceptable contact with their surroundings. The approach expands how robots reason about physical interaction, allowing for more natural and effective movement in environments where contact is unavoidable.
Other work focuses on making robot learning more data-efficient and scalable. “AutoFocus-IL: VLM-Based Saliency Maps for Data-Efficient Visual Imitation Learning Without Extra Human Annotations” leverages VLMs to generate saliency maps that highlight the most relevant parts of a scene, reducing the need for manual labeling while improving imitation learning performance. Similarly, “PSALM-V: Automating Symbolic Planning in Interactive Visual Environments with Large Language Models” uses LLMs to automate high-level symbolic planning, enabling robots to carry out more autonomous reasoning in visually complex, interactive settings.
USC papers also explore advanced learning systems that improve how robots evaluate success and adapt to new tasks. “Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons” introduces a framework for learning reward models by comparing movement trajectories, helping robots better understand successful behavior across tasks. Meanwhile, “HAND Me the Data: Fast Robot Adaptation via Hand Path Retrieval” accelerates adaptation by allowing robots to learn from retrieved human hand motion patterns. “Robot Learning from a Physical World Model,” further advances how robots perceive structure, model physical reality and translate visual input into precise, real-time actions.
Building Smarter Robots with Generative AI and Data-Driven Robot Learning
USC researchers are increasingly turning to generative AI to help robots learn faster, perceive more accurately and adapt to environments that are difficult to capture through real-world data alone. Several papers presented at ICRA this year explore how diffusion models and synthetic data generation can expand the range and quality of training data available for robotic systems.
In “AnchorDream: Repurposing Video Diffusion for Embodiment-Aware Robot Data Synthesis,” researchers repurpose video diffusion models to generate synthetic robotic training data that is aware of a robot’s physical form and perspective. By producing embodiment-aware visual experiences, the approach helps bridge the gap between simulated learning environments and real-world deployment, improving how robots generalize from synthetic data.
Other work focuses on expanding and diversifying robotic datasets through augmentation techniques. “ROPA: Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation” introduces a method for generating synthetic robot poses to enhance training data for bimanual systems, enabling more robust two-armed manipulation learning using RGB-D inputs. Together, these approaches help address the challenge of collecting large-scale, high-quality datasets for complex robotic tasks.
USC research also tackles difficult perception problems using generative modeling. “Diffusion Knows Transparency: Repurposing Video Diffusion for Transparent Object Depth and Normal Estimation” leverages video diffusion models to estimate depth and surface properties of transparent objects, which are often difficult for traditional sensors to interpret. By improving how robots perceive materials like glass and plastic, the work helps close critical gaps in real-world robotic perception.
Autonomous Robots for Space and Off-World Environments
Robots are the scouts of humanity, able to reach the most distant and hazardous regions of space; they will be essential partners in enabling humans to live and work beyond Earth. That priority is reflected by USC Viterbi’s Department of Astronautical Engineering (ASTE), where space robotics is a growing area of research. ASTE is represented at ICRA 2026 by Keenan Albee, assistant professor of astronautical engineering and aerospace and mechanical engineering, who specializes in autonomous systems that can operate in extreme and unmapped environments.
In addition to serving as organizer and program chair for the “Intelligent Space Robotics and Systems” workshop, Albee has co-authored two papers presented at ICRA. “Safe Payload Transfer with Ship-Mounted Cranes: A Robust Model Predictive Control Approach” presents a new control framework for enabling robotic crane systems to safely operate under the highly unstable conditions created by ocean motion, demonstrating how autonomous systems can maintain precision and safety even in unpredictable environments. “CRESCENT: Collision-Free Highly Constrained Trajectory Optimization for Driving on the Moon” introduces a real-time trajectory optimization system developed for NASA’s upcoming CADRE lunar rover mission, allowing rovers to independently navigate cluttered, unmapped terrain with minimal human supervision. The papers tackle key challenges for the space industry, seeking to advance the perception, planning and control capabilities needed for future robotic exploration missions and long-duration operations beyond Earth.
USC-Affiliated Papers
(USC authors bolded)
Satyajeet Das, Darren Chiu, Zhehui Huang, Lars Lindemann, Gaurav Sukhatme
Refinery: Active Fine-Tuning and Deployment-Time Optimization for Contact-Rich Policies
Bingjie Tang, Iretiayo Akinola, Jie Xu, Bowen Wen, Dieter Fox, Gaurav Sukhatme, Fabio Ramos, Abhishek Gupta, Yashraj Narang
ROPA: Synthetic Robot Pose Generation for RGB-D Bimanual Data Augmentation
Jason Chen, I-Chun Arthur Liu, Gaurav Sukhatme, Daniel Seita
Learning Geometry-Aware Nonprehensile Pushing and Pulling with Dexterous Hands
Yunshuang Li, Yiyang Ling, Gaurav Sukhatme, Daniel Seita
IMPACT: Intelligent Motion Planning with Acceptable Contact Trajectories via Vision-Language Models
Yiyang Ling, Karan Owalekar, Oluwatobiloba Adesanya, Erdem Bıyık, Daniel Seita
V-MORALS: Visual Morse Graph-Aided Discovery of Regions of Attraction in a Learned Space
Faiz Aladin, Ashwin Balasubramanian, Lars Lindemann, Daniel Seita
SCOOP’D: State-based Sim2Real Generative Policy for Generalizable Mixed-Liquid-Solid Scooping
Kuanning Wang, Yongchong Gu, Yuqian Fu, Zeyu Shangguan, Sicheng He, Xiangyang Xue, Yanwei Fu, Daniel Seita
OCRA: Object-Centric Learning with 3D and Tactile Priors for Human-to-Robot Action Transfer
Kuanning Wang, Ke Fan, Yuqian Fu, Siyu Lin, Hu Luo, Daniel Seita, Yanwei Fu, Yu-Gang Jiang, Xiangyang Xue
Preference-Conditioned Reinforcement Learning for Space-Time Efficient Online 3D Bin Packing
Nikita Sarawgi, Omey M. Manyar, Fan Wang, Thinh H. Nguyen, Daniel Seita, Satyandra K. Gupta
Litian Gong, Fatemeh Bahrani, Yutai Zhou, Amin Banayeeanzade, Jiachen Li, Erdem Bıyık
HAND Me the Data: Fast Robot Adaptation via Hand Path Retrieval
Matthew Hong, Anthony Liang, Kevin Kim, Harshitha Rajaprakash, Jesse Thomason, Erdem Bıyık, Jesse Zhang
Jesse Zhang, Marius Memmel, Kevin Kim, Dieter Fox, Jesse Thomason, Fabio Ramos, Erdem Bıyık, Abhishek Gupta, Anqi Li
Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons
Anthony Liang, Yigit Korkmaz, Jiahui Zhang, Minyoung Hwang, Abrar Anwar, Sidhant Kaushik, Aditya Shah, Alex S. Huang, Luke Zettlemoyer, Dieter Fox, Yu Xiang, Anqi Li, Andreea Bobu, Abhishek Gupta, Stephen Tu, Erdem Bıyık, Jesse Zhang
Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation
Zhenyu Zhao, Hongyi Jing, Xiawei Liu, Jiageng Mao, Abha Jha, Hanwen Yang, Rong Xue, Sergey Zakharov, Vitor Guizilini, Yue Wang
SeFA-Policy: Fast and Accurate Visuomotor Policy Learning with Selective Flow Alignment
Rong Xue, Jiageng Mao, Mingtong Zhang, Yue Wang
AnchorDream: Repurposing Video Diffusion for Embodiment-Aware Robot Data Synthesis
Junjie Ye, Rong Xue, Basile Van Hoorick, Pavel Tokmakov, Muhammad Zubair Irshad, Yue Wang, Vitor Guizilini
Robot Learning from a Physical World Model
Jiageng Mao, Sicheng He, Hao-Ning Wu, Yang You, Shuyang Sun, Zhicheng Wang, Yanan Bao, Huizhong Chen, Leonidas Guibas, Vitor Guizilini, Howard Zhou, Yue Wang
Shaocong Xu, Songlin Wei, Qizhe Wei, Zheng Geng, Hong Li, Licheng Shen, Qianpu Sun, Shu Han, Bin Ma, Bohan Li, Chongjie Ye, Yuhang Zheng, Nan Wang, Saining Zhang, Hao Zhao
Eventually Optimal and Scalable Multi-Agent Planning for Block Cave Mining
Christopher Leet, Paolo Forte, Uwe Köckemann, Henrik Andreasson, Sven Koenig
Learning to Drive Anywhere with Model-Based Reannotation
Noriaki Hirose, Lydia Ignatova, Kyle Stachowicz, Catherine Glossop, Sergey Levine, Dhruv Shah
One Filter to Deploy Them All: Robust Safety for Quadrupedal Navigation in Unknown Environments
Albert Lin, Shuang Peng, Somil Bansal
Safety Evaluation of Motion Plans Using Trajectory Predictors As Forward Reachable Set Estimators
Kaustav Chakraborty, Zeyuan Feng, Sushant Veer, Apoorva Sharma, Wenhao Ding, Sever Topan, Boris Ivanovic, Marco Pavone, Somil Bansal
UMBRELLA: Uncertainty-Aware Multi-Robot Reactive Coordination under Dynamic Temporal Logic Tasks
Qisheng Zhao, Meng Guo, Hengxuan Du, Lars Lindemann, Zhongkui Li
Chenrui Tie, Shengxiang Sun, Yudi Lin, Yanbo Wang, Zhongrui Li, Zhouhan Zhong, Jinxuan Zhu, Yiman Pang, Haonan Chen, Junting Chen, Ruihai Wu, Lin Shao
Safe Payload Transfer with Ship-Mounted Cranes: A Robust Model Predictive Control Approach
Ersin Das, William A. Welch, Patrick Spieler, Keenan Albee, Aurelio Noca, Jeffrey Edlund, Jonathan Becktor, Thomas Touma, Jessica Todd, Sriramya Bhamidipati, Stella Kombo, Maira Saboia, Anna Sabel, Grace Lim, Rohan Thakker, Amir Rahmani, Joel W. Burdick
CRESCENT: Collision-Free Highly Constrained Trajectory Optimization for Driving on the Moon
Abhishek Cauligi, Keenan Albee, Jean-Pierre De La Croix, Roland Brockers
STATE-NAV: Stability-Aware Traversability Estimation for Bipedal Navigation on Rough Terrain
Ziwon Yoon, Lawrence Y. Zhu, Jingxi Lu, Lu Gan, Ye Zhao
Bio-Inspired Tail Oscillation Enables Fast Crawling on Deformable Granular Terrains
Shipeng Liu, Meghana Sagare, Shubham Patil, Feifei Qian
PSALM-V: Automating Symbolic Planning in Interactive Visual Environments with Large Language Models
Wang Bill Zhu, Miaosen Chai, Ishika Singh, Robin Jia, Jesse Thomason
Real-Time Seam Tracking for Robotic Welding Via Registration-Based Deformation Estimation
Surag Balajepalli, Amrish Baskaran, Shounak Naik, Christopher Eubel, Abolfazl Meyarian, Andong Dai, Pradeep Rajendran, Katsu Yamane, Alex Trazkovich
SSQA: Sibling-Selective Quadtree Attention for Hierarchical Modeling in Perception Tasks
Yufan Chen, Arnav Bali, Angela Liu, Laura Zheng, Ming C. Lin
Javier Borquez, Luke Raus, Yusuf Umut Ciftci, Somil Bansal
Published on June 1st, 2026
Last updated on June 3rd, 2026