Brain-Inspired Robotics Laboratory’s Post

Hey everyone! 🌟 We want to announce our participation in the upcoming IEEE #BioRob2024 conference in Heidelberg, taking place from September 1st to 4th . Our team will be contributing with two oral presentations and organizing a workshop. 📄 𝗢𝗿𝗮𝗹 𝗣𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻𝘀 [September 2, 2024 @ 5-6.30pm - Room HS7] [Track: Soft Robotics] • 𝘉𝘦𝘩𝘢𝘷𝘪𝘰𝘳 𝘊𝘭𝘰𝘯𝘪𝘯𝘨 𝘧𝘳𝘰𝘮 𝘖𝘣𝘴𝘦𝘳𝘷𝘢𝘵𝘪𝘰𝘯𝘴 𝘸𝘪𝘵𝘩 𝘋𝘰𝘮𝘢𝘪𝘯 𝘔𝘢𝘱𝘱𝘪𝘯𝘨 𝘧𝘰𝘳 𝘵𝘩𝘦 𝘊𝘰𝘯𝘵𝘳𝘰𝘭 𝘰𝘧 𝘚𝘰𝘧𝘵 𝘙𝘰𝘣𝘰𝘵𝘴 by Beatrice Tosi, Muhammad Sunny Nazeer, Egidio Falotico The designs of soft robots often draw inspiration from biological systems, even though their morphologies and actuation mechanisms often differ significantly. Reproducing bio-inspired movements on a corresponding platform necessitates in-depth knowledge of these mechanisms. This paper introduces a simplified approach to tackle this challenge in an imitation learning-based problem setting. • 𝘛𝘰𝘸𝘢𝘳𝘥𝘴 𝘐𝘯𝘵𝘦𝘳𝘱𝘳𝘦𝘵𝘢𝘣𝘭𝘦 𝘝𝘪𝘴𝘶𝘰-𝘛𝘢𝘤𝘵𝘪𝘭𝘦 𝘗𝘳𝘦𝘥𝘪𝘤𝘵𝘪𝘷𝘦 𝘔𝘰𝘥𝘦𝘭𝘴 𝘧𝘰𝘳 𝘚𝘰𝘧𝘵 𝘙𝘰𝘣𝘰𝘵 𝘐𝘯𝘵𝘦𝘳𝘢𝘤𝘵𝘪𝘰𝘯𝘴 by Enrico Donato, Thomas George Thuruthel, Egidio Falotico Autonomous systems struggle with navigating unpredictable environments and interacting with external objects, relying on complex perception capabilities. Our work develops a multi-modal perception model for soft robots using a generative approach, combining proprioceptive and visual data to predict contact forces. We provide tools to interpret this model, clarifying the fusion and prediction processes across sensory inputs, as well as cross-modality generation. Pre-print: https://lnkd.in/dR4W9DBx 🛠️ 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽: 𝘚𝘦𝘯𝘴𝘰𝘳𝘪𝘮𝘰𝘵𝘰𝘳 𝘙𝘦𝘱𝘳𝘦𝘴𝘦𝘯𝘵𝘢𝘵𝘪𝘰𝘯: 𝘈𝘥𝘷𝘢𝘯𝘤𝘦𝘮𝘦𝘯𝘵𝘴 𝘪𝘯 𝘓𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘔𝘰𝘥𝘦𝘭𝘴 𝘧𝘰𝘳 𝘙𝘰𝘣𝘰𝘵𝘴 [September 1, 2024 @ 2-6pm - Room HS7] We will explore recent advancements in AI and computational neuroscience that enhance robotic capabilities in unstructured environments. Topics include deep learning architectures, probabilistic models, and brain-inspired frameworks such as spiking neural networks. We will delve into cognitive architectures for motor control, inspired by human cognition, which provide efficient and adaptable frameworks for learning and executing motor commands. Organizers: Egidio Falotico, Elisa Donati, Matej Hoffmann Confirmed speakers: Pablo Lanillos, Chiara Bartolozzi, Yulia Sandamirskaya, Mehdi Khamassi, Martin Butz 𝘋𝘦𝘢𝘥𝘭𝘪𝘯𝘦 𝘧𝘰𝘳 𝘤𝘰𝘯𝘵𝘳𝘪𝘣𝘶𝘵𝘪𝘰𝘯𝘴: 𝘈𝘶𝘨𝘶𝘴𝘵 9, 2024 @ 5𝘱𝘮 More info: https://lnkd.in/dZW84PTt We are looking forward to seeing familiar faces and meeting new ones. If you are going to be at #BioRob2024, come join us for our presentations and workshop. Let’s chat, share ideas, and maybe even spark some new, exciting collaborations! See you there! #Innovation #Research #Biomedical #Robotics #Biomechatronics #Conference

📃Scientific paper: Humanising robot-assisted navigation Abstract: Robot-assisted navigation is a perfect example of a class of applications requiring flexible control approaches. When the human is reliable, the robot should concede space to their initiative. When the human makes inappropriate choices the robot controller should kick-in guiding them towards safer paths. Shared authority control is a way to achieve this behaviour by deciding online how much of the authority should be given to the human and how much should be retained by the robot. An open problem is how to evaluate the appropriateness of the human's choices. One possible way is to consider the deviation from an ideal path computed by the robot. This choice is certainly safe and efficient, but it emphasises the importance of the robot's decision and relegates the human to a secondary role. In this paper, we propose a different paradigm: a human's behaviour is correct if, at every time, it bears a close resemblance to what other humans do in similar situations. This idea is implemented through the combination of machine learning and adaptive control. The map of the environment is decomposed into a grid. In each cell, we classify the possible motions that the human executes. We use a neural network classifier to classify the current motion, and the probability score is used as a hyperparameter in the control to vary the amount of intervention. The experiments collected for the paper show the feasibility of the idea. A qualitative evaluation, done by surveying the ... Continued on ES/IODE ➡️ https://etcse.fr/h3ZE ------- If you find this interesting, feel free to follow, comment and share. We need your help to enhance our visibility, so that our platform continues to serve you.

Excited to share that my first article is now live on Medium! 🚀 In this piece, I dive deep into the fascinating world of robotics and AI, focusing on the crucial role of optimizing deep learning models for robotic systems. As we push the boundaries of what's possible with technology, understanding how to tailor AI models to specific hardware constraints becomes increasingly important. 🤖💡 I've unpacked the complexities surrounding tools such as NVIDIA's TensorRT and Intel's OpenVINO toolkit. If you're fascinated by the advancements in robotics, an AI developer, or simply intrigued by how deep learning is shaping our technological landscape, this article has something for you. Check it out and let me know your thoughts! Feedback, discussions, and shares are highly encouraged. 🔍🌐 #Robotics #ArtificialIntelligence #DeepLearning #Technology #Innovation

📃Scientific paper: MoEmo Vision Transformer: Integrating Cross-Attention and Movement Vectors in 3D Pose Estimation for HRI Emotion Detection Abstract: Emotion detection presents challenges to intelligent human-robot interaction (HRI). Foundational deep learning techniques used in emotion detection are limited by information-constrained datasets or models that lack the necessary complexity to learn interactions between input data elements, such as the the variance of human emotions across different contexts. In the current effort, we introduce 1) MoEmo (Motion to Emotion), a cross-attention vision transformer (ViT) for human emotion detection within robotics systems based on 3D human pose estimations across various contexts, and 2) a data set that offers full-body videos of human movement and corresponding emotion labels based on human gestures and environmental contexts. Compared to existing approaches, our method effectively leverages the subtle connections between movement vectors of gestures and environmental contexts through the use of cross-attention on the extracted movement vectors of full-body human gestures/poses and feature maps of environmental contexts. We implement a cross-attention fusion model to combine movement vectors and environment contexts into a joint representation to derive emotion estimation. Leveraging our Naturalistic Motion Database, we train the MoEmo system to jointly analyze motion and context, yielding emotion detection that outperforms the current state-of-the-art. ;Comment: IEEE/RSJ International Conference on Intelligent Robots (IROS), Detroit, Michigan Continued on ES/IODE ➡️ https://etcse.fr/fjh ------- If you find this interesting, feel free to follow, comment and share. We need your help to enhance our visibility, so that our platform continues to serve you.

Researchers are pushing the boundaries of artificial intelligence (AI) by developing robots that can simulate "dreaming." This process involves using simulations or imagined scenarios to help robots learn and adapt without direct real-world experience. By incorporating virtual environments, robots can practice tasks, refine decision-making, and explore hypothetical situations, improving their performance in real-world applications. This approach mimics how humans and animals process experiences during sleep to enhance learning. The research highlights the potential for "machine dreaming" to make AI more autonomous and efficient in learning new tasks. #AI Source: https://lnkd.in/eFvCBUtm

The legendary sci-fi writer Philip K. Dick wrote a book called “Do Androids Dream of Electric Sheep?” (which - in addition to being the basis for the movie Blade Runner - has to be one of the coolest titles ever). Massachusetts Institute of Technology researchers are answering that question - sort of - with their new LucidSim platform: inspired by human dreaming, it trains robot dogs using GenAI simulated "dreams." Instead of running countless physical tests, the robot trains in a virtual environment, mastering parkour and more through generative AI. The result? Faster learning, lower costs, and a new frontier for robotics innovation. Kind of like Neo learning martial arts in The Matrix, to use another sci-fi analogy. To me this is a perfect example of the philosophy we follow here at BlueSkyAI: solving real-world challenges by blending creativity and computational power. In AI, whether you're optimizing marketing campaigns or teaching a robot to navigate rough terrain, the principle is the same: human ingenuity can leverage AI for maximum results. #matrix #ai #genai #mit #robotics #andriods #tech https://lnkd.in/e7KyJBmK

"Scientists create AI models that can talk to each other and pass on skills with limited human input" (https://lnkd.in/ek72GBvJ) JK--Wow. Transfer learning as a conversational process that bridges cognitive, sensory, and motor processing domains. Robotics inflection point?

What Will Happen if Robotics and AI ML Gets Combined? The integration of Robotics and Artificial Intelligence (AI), including Machine Learning (ML), is a monumental step towards revolutionizing the future. Imagine a world where robots can not only perform repetitive tasks but also think, adapt, and learn like humans. This convergence of Robotics and AI is already underway, transforming industries and reshaping the global economy. But what exactly happens when these two powerful technologies are combined? In this blog, we delve into the synergy of Robotics and AI, exploring its implications, benefits, and the potential challenges that lie ahead. Read More :-https://lnkd.in/dEggVKNb #leansimcorp #beawarechannel #blogs #blogger #education #learning #ENGINEERING #mechanical #industrial

Marvin Minsky: A Visionary Who Bridged AI, Robotics, and Engineering Ever wondered what it takes to revolutionize not just one field, but multiple? Marvin Minsky, co-founder of MIT's Artificial Intelligence Lab, faced immense challenges while shaping the future of AI and robotics. Take the Minsky Arm, one of the first computer-controlled robotic arms. Developing it in the 1960s meant grappling with primitive computing power, limited sensors, and skepticism about AI's potential. Yet, Minsky's relentless pursuit turned it into a foundational tool that paved the way for modern robotics. At MIT, Minsky’s visionary leadership helped establish AI as a discipline. He proposed the Society of Mind, a groundbreaking theory that intelligence arises from smaller, simpler processes working together. His lab became a hub for innovation, inspiring technologies in natural language processing, machine learning, and robotics. Today, Minsky’s legacy is embedded in modern engineering and robotics. From autonomous machines to AI-powered systems, his contributions continue to shape industries, reminding us that progress often comes through perseverance in the face of doubt. Minsky’s journey isn’t just a story of inventions—it’s a testament to pushing the boundaries of what’s possible. What boundaries are you ready to challenge? #Innovation

RoboCasa: Revolutionizing Robot Training 🦾 Researchers have unveiled RoboCasa, a simulation framework designed to train general-purpose robots in everyday settings. RoboCasa offers access to thousands of 3D assets across 150+ object categories and dozens of interactive furniture and appliances. Generative AI tools enhance its realism and diversity, utilizing text-to-3D models for objects and text-to-image models for environments. Key findings show that large-scale imitation learning with synthetic data demonstrates promising scalability for real-life robotic tasks. This research, detailed on AriXiv by teams from UT Austin and NVIDIA Research, addresses the scarcity of diverse robotic training data compared to the abundant data available for computer vision and NLP. Built on RoboSuite's foundation, RoboCasa provides 120 realistic scenes, over 2,500 3D objects, and supports 100 tasks with a dataset of 100,000+ trajectories. Generative AI tools play a crucial role, creating diverse, realistic simulations. Despite improvements, challenges remain in transferring learned policies to real-world environments and increasing task complexity. Future work aims to enhance simulation fidelity and bridge the gap between simulated and real-world applications, unlocking new potential for robotic training and deployment. #ai #robotics #simulation #generativeai #robocasa #tech