Hello from all of us at Hello Robot!
Last month we were thrilled to unveil Stretch 4 – our best robot yet! We incorporated years of community feedback into a ground-up redesign of the platform. Major highlights include an all-new omnidirectional base, 360 degree sensor head with two hemispherical 3D LiDAR sensors, greatly increased battery life and cycle life, and new available options for an onboard Jetson Orin, parallel jaw gripper, and autonomous docking. Visit the redesigned hello-robot.com for more details.
This Stretch Community News showcases advancements in embodied AI, including new frameworks for multi-robot collaboration (IndoorR2X), generalized indoor navigation (One RING), and improved motion planning for cluttered environments (MOPS-PRM). You will also find research focused on making domestic robots more adaptable using LLMs, enhancing robustness in language-guided grasping, and exploring real-world applications in areas like contactless pollination and assistive technology for aging adults.
Read on for more details! And if you’d like your work featured in a future newsletter, we’d love to hear from you. Drop us a note at community@hello-robot.com.
Cheers,
Aaron Edsinger
CEO – Hello Robot
In May we announced the launch of Stretch 4, the newest generation of our mobile manipulator robot platform. This IEEE Spectrum article explores the thinking behind Stretch 4’s design, highlights its new capabilities, and discusses how the platform is being developed to support both research and piloting real-world applications.
Researchers from Carnegie Mellon University, Toyota Research Institute, and Waseda University recently introduced IndoorR2X, a framework that improves collaboration between indoor robots using robot-to-robot communication and smart building sensors. By combining data from mobile robots, cameras, and IoT devices, and leveraging Large Language Models (LLMs) for task planning, the system improved the efficiency and reliability of multi-robot coordination in indoor environments.
Researchers from Lawrence Technological University are exploring how foundation models and Large Language Models (LLMs) can help domestic service robots become more adaptable in home environments. Tested using mobile phone data and the Stretch 2 Mobile Manipulator, the system identifies meaningful changes in a home and determines appropriate actions, helping robots operate more proactively and efficiently.
Researchers from Khalifa University, the University of Windsor, and Thapar Institute of Engineering and Technology developed a contactless robotic pollination system for greenhouse agriculture. Using real-time flower detection and pose estimation, the system, integrated with the Stretch platform and a custom “Air Pollenmatic” tool, achieved strong pollination success rates while helping reduce flower damage and disease transmission.
A recent Barron’s article highlights the growing popularity of university-affiliated retirement communities, where older adults can access lifelong learning, cultural programs, and intergenerational experiences alongside traditional care services. The story features programs connected to the University of Illinois Urbana-Champaign and Arizona State University and also highlights research involving the Stretch robot as a tool to support healthy aging and independent living.
Researchers from Rice University and the Ken Kennedy Institute developed MOPS-PRM, a motion-planning system designed to help robots operate more effectively in cluttered environments. By maintaining awareness of multiple people and objects during navigation, the system improved detection and tracking performance while preserving planning efficiency, advancing the development of safer and more perceptive indoor robots.
A recent feature from the Georgia Institute of Technology highlights how students are using the Stretch robot through the Stretch Robot Pitch Competition to develop assistive robotics concepts for older adults and people aging with disabilities. Supported by organizations including Center for Inclusive Design and Innovation, TechSAge, AI-CARING, and Hello Robot, the program encourages innovative solutions for daily living challenges, with winning teams advancing their ideas in Georgia Tech’s Aware Home research environment.
Researchers from the Australian Institute for Machine Learning at Adelaide University developed a new language-guided robotic grasping approach that helps robots recover from failed actions. Using an execution monitoring system called Watchdog and an “agentic loop” for decision-making, the framework improved robustness by detecting failures, deciding when to retry, and requesting clarification when needed.
Researchers from the Allen Institute for AI recently introduced One RING, a generalized indoor navigation system that helps robots navigate unfamiliar environments across multiple platforms, including the Stretch RE-1. Using a shared policy rather than robot-specific models, the system demonstrated strong generalization across different robots and environments, highlighting progress toward more adaptable embodied AI systems.
Researchers from Johns Hopkins University, Lambda, and the University of Cambridge developed 3D-Belief, a generative world model that helps robots maintain a 3D understanding of partially observed environments. By continuously updating its spatial memory as new observations are gathered, 3D-Belief improved scene understanding and task performance in both simulation and real-world navigation experiments, advancing robotic navigation under uncertainty.
Researchers recently introduced AbsVLA, a framework designed to make Vision–Language–Action (VLA) robotic systems more robust when visual conditions or instructions change. By using object-centric abstract representations that focus on task-relevant spatial information, AbsVLA improved performance on the LIBERO benchmark and demonstrated zero-shot transfer from simulation to a real robot with a different embodiment.