#Cognitive Robotics

[compneuropapers]

Interesting Reviews for Week 11, 2025

A global perspective on AI innovation and effective use in the research lab. Jeng, A. C., Sibley, I. J., & Bale, T. L. (2024). Neuroscience, 560, 106–108.

Advances and challenges in the development of visual prostheses. Fernandez, E., & Robles, J. A. (2024). PLOS Biology, 22(10), e3002896.

Attention Deficit Hyperactivity Disorder, Cannabis Use, and the Endocannabinoid System: A Scoping Review. Ryan, J. E., Fruchtman, M., Sparr‐Jaswa, A., Knehans, A., & Worster, B. (2024). Developmental Psychobiology, 66(7).

Understanding the sense of self through robotics. Prescott, T. J., Vogeley, K., & Wykowska, A. (2024). Science Robotics, 9(95).

[neosciencehub]

Intelligent Robotic Teammate for Defence Soldiers

Intelligent Robotic Teammate for Defence Soldiers @neosciencehub #neosciencehub #science #Robotics #Defense #AI #MilitaryApplications #AutonomousSystems #CognitiveRobotics #TechnologicalInnovations #research #industry #IITDelhi #risk

IIT Delhi Pioneers Development of a Teammate A team of researchers from the Indian Institute of Technology Delhi, in collaboration with the Defence-Industry-Academia Centre of Excellence (DIA-CoE) and the DRDO Centre for Artificial Intelligence & Robotics (CAIR), has made a significant breakthrough in robotic technology at the IInvenTive 2024 R&D fair. The project, led by Prof. Rohan Paul and…

[art-for-a-reason]

The Electric State

Just watched The Electric State on Netflix. Not going to say too much about it, but I think the critics completely and utterly missed the point(s) of this movie. It was poignant and topical and well written. It was also fun and interesting. It did not spoon-feed the viewer, however, and there were times when I was like, "is this an AI apologist vehicle?" To which I can definitively answer, "No." Just view it through the lens of what we as a nation are going through right now, and it'll find the appropriate spots to settle.

[chasm-connected]

i spent literally all of my currency in pixel cats on him. i have like some pennies left. worth it.

[sunsetzer]

It's truly amazing what having an unkind boss does to your psyche. I still have dreams where I'm chastised for not being good enough at my job and replaced with someone else, and it's been over a year since I got a new manager.

[dideki]

university.. university leave me alone

[sillimancer]

yknow when my therapist closed last week's session with a "so next week is the election please talk to me before considering dying" talk I kinda brushed it off but now it's two days away and there it is, that funny feeling once again.

[libraryofbaxobab]

August 20, 2024:

This didn't grab me like "The Tusks Of Extinction" did, even though it has very similar environmentalist and, more specifically, animal cognition themes (and the same author). But octopuses are really cool, and their intelligence is creepy and exhilarating to think about. That's pretty much it, I guess.

5/10 #WhatsKenyaReading

[jcmarchi]

How NVIDIA Isaac GR00T N1 Is Redefining Humanoid Robotics

New Post has been published on https://thedigitalinsider.com/how-nvidia-isaac-gr00t-n1-is-redefining-humanoid-robotics/

How NVIDIA Isaac GR00T N1 Is Redefining Humanoid Robotics

For decades, scientists and engineers have worked to create humanoid robots capable of walking, talking, and interacting like humans. While significant progress has been made, building robots that can adapt to new environments or learn new skills has remained a complex and costly challenge. NVIDIA is addressing this with Isaac GR00T N1, the world’s first open and customizable foundation model for humanoid robot reasoning and skills. This innovative model equips robots with the ability to think critically, reason through complex scenarios, and adapt to new challenges. This article explores NVIDIA’s innovation, detailing GR00T N1’s features and its impact on humanoid robotics.

The Current State of Humanoid Robotics

Humanoid robotics has advanced considerably in recent years. They can walk across uneven terrain, carry on basic conversations, and handle tasks like assembling products in controlled environments. Companies like Boston Dynamics have demonstrated robots that can dance or perform acrobatics. However, despite all these advancements, these robots face limitations when confronted with tasks outside their specific programming. For example, a robot designed to stack boxes in a warehouse may struggle to sort items in a cluttered storeroom or switch tasks without extensive reprogramming. Primarily, building a humanoid robot capable of handling diverse tasks required starting from scratch each time, a process that could take months or even years.

A Foundation Model for Humanoid Robotics

The Isaac GR00T N1 is a foundation model specifically designed for humanoid robots. It provides a pre-built framework for essential functions like perception and movement, eliminating the need to develop these core capabilities from scratch. This simplifies the robot-building process, which previously demanded expertise in fields like mechanical engineering and AI programming, along with significant financial resources. Developers can now take GR00T N1 and customize it for specific tasks, reducing both time and cost. This accessibility and flexibility could drive wider adoption, enabling these robots to move from research labs to real-world applications.

Thinking Like Humans: A Dual-System Design

GR00T N1 employs dual-system design inspired by human cognition. According to dual process theory, humans think in two modes: fast and instinctive (like reflexes) and slow and deliberate (like planning). Following this cognitive model, GR00T N1 is equipped with both System 1 and System 2. System 1 enables GR00T to handle quick reactions, such as dodging obstacles or catching moving objects, similar to human reflexes. On the other hand, System 2 allows GR00T to process more complex tasks, like processing instructions, analyzing visual data, or planning multi-step actions such as organizing a messy room. By combining these systems, GR00T N1-powered robots can tackle diverse challenges with human-like flexibility. For instance, a robot could pick up scattered items, decide where they belong, and navigate unexpected barriers, all while adapting in real time.

Training GR00T N1

Training GR00T to think and move like a human requires vast amounts of data, which can be slow and expensive to collect in real-world settings. NVIDIA addresses this with the Isaac GR00T Blueprint, a tool that generates synthetic motion data in virtual environments. Starting with a small set of human demonstrations, the blueprint can produce large datasets quickly. In one example, NVIDIA created 780,000 synthetic trajectories—equivalent to 6,500 hours of human effort—in just 11 hours. Combining this synthetic data with real-world data improved GR00T N1’s performance by 40% compared to using real data alone. This method speeds up learning, enhances adaptability, and refines skills without relying heavily on physical trials.

Impact on Humanoid Robotics

Building a robot and its AI from scratch has traditionally been a slow and costly endeavor. GR00T N1 changes this by providing a model pre-trained in reasoning and movement, allowing developers to focus on customization. This could accelerate deployment in industries like manufacturing, logistics, and healthcare, where adaptable solutions are increasingly needed. A GR00T N1-powered robot might move materials, pack goods, or assist with patient care, switching roles as required.

NVIDIA has made GR00T N1 freely available to the global robotics community, unlike proprietary systems that restrict access. This openness allows startups, researchers, and large companies to download, modify, and adapt it, enabling smaller teams with limited resources to innovate alongside industry leaders.

GR00T N1 processes multiple input types, such as language and visual data, allowing robots to interpret spoken commands, recognize objects, and adapt to changing environments. This versatility is critical for humanoid robots operating in the unpredictable reality of human spaces. Unlike traditional robots built for repetitive tasks in structured settings, GR00T N1-powered robots excel in dynamic roles—like healthcare assistance or logistics management—where flexibility and natural interaction are key.

GR00T in Action: Real-World Applications

Companies like Boston Dynamics, Agility Robotics, and 1X Technologies are testing GR00T N1. In manufacturing, these robots can assemble parts or sort packages and adjust to production changes. Their ability to switch tasks easily fits factories needing flexibility.

In healthcare, they could lift patients from beds to wheelchairs using voice guidance from nurses. They might also assist elderly people by fetching items and talking naturally. GR00T N1’s understanding of language and context makes these interactions more natural and human-like. For example, 1X Technologies’ NEO Gamma robot used GR00T N1 to autonomously tidy up a home. It assessed the space, decided what to do, like picking up toys or fixing a table, and acted on its own. This suggests how GR00T-powered robots can become household helpers, aiding with chores or supporting those with mobility issues.

NVIDIA’s Future Plans for Advancing Humanoid Robotics

Besides GR00T, NVIDIA is also working with Google DeepMind and Disney Research to develop a physics engine, Newton, for humanoid robotics. This open-source tool enables robotics developers to simulate how robots move and interact with their surroundings. It can integrate with platforms like MuJoCo and NVIDIA Isaac Lab and help test robots virtually before they step into reality. This development will further lower costs, cut risks, and speed up robot development.

The Bottom Line

NVIDIA’s Isaac GR00T N1 offers a significant advancement in humanoid robotics by providing a customizable foundation for reasoning and movement. Its dual-system design allows robots to quickly respond to changes and handle complex tasks, adapting to various environments. By using synthetic data for training, the model reduces both development time and costs. Offering GR00T N1 as an open model encourages innovation across industries such as manufacturing, healthcare, and logistics. Early implementations show the model’s potential to enhance flexibility and efficiency in real-world applications.

[compneuropapers]

Interesting Reviews for Week 27, 2023

Attractor and integrator networks in the brain. Khona, M., & Fiete, I. R. (2022). Nature Reviews Neuroscience, 23(12), 744–766.

Learning robotic navigation from experience: principles, methods and recent results. Levine, S., & Shah, D. (2023). Philosophical Transactions of the Royal Society B: Biological Sciences, 378(1869).

Querying hippocampal replay with subcortical inputs. Peyrache, A. (2022). Current Opinion in Neurobiology, 77, 102645.

Attentional biases in human anxiety. Valadez, E. A., Pine, D. S., Fox, N. A., & Bar-Haim, Y. (2022). Neuroscience & Biobehavioral Reviews, 142, 104917.

[thekavach]

How robotics can enhance problem solving skills in young learners

Robotics in education provides young learners with an interactive platform to strengthen their problem-solving abilities. Through designing, building, and programming robots, students cultivate logical thinking, creativity, and resilience. They develop a structured approach to challenges, embrace trial and error, and enhance collaboration skills. This hands-on learning experience bridges the gap between theory and practice while nurturing computational thinking, equipping them for future technological advancements. Visit here for more query https://kavach.org.in/how-robotics-can-enhance-problem-solving-skills-in-young-learners/

[a-girl-called-bob]

i don't think you can expect robot girls to understand code any more that you can expect random carbon-based people to understand Flemish

i don't think you can expect robot girls to understand code any more that you can expect random carbon-based people to understand nucleotide sequences

[stealthcake]

AI Makes You Stupid

I freaking knew it you guys.

Microsoft Study Finds AI Makes Human Cognition “Atrophied and Unprepared”

So not to be all conspiratorial, but is this why corporations are pushing AI so hard? Because they are really pushing it. I can't do anything online or on my computer without having to make an effort to avoid AI. It's getting beyond the point where it feels like a sales pitch or fake added value. It feels like Big Tech is trying to possess and commodify what little remains of real human cognition and communication.

[divinekangaroo]

also I keep imagining that I'm going to go massively off script and suddenly start talking about the nightclub dancer I picked up when I was 19 for an extended one night stand

[tilbageidanmark]

[kanguin]

Like. Guys. It's the art theft and plagiarism that's the problem. When AI bros say they couldn't make their tools and programs without taking from others, they are LYING. The truth is that these tools and programs have been made ethically by studios for years, and still are! It just costs more time and money to do it the ethical way, and they don't like that! Convincing everyone that theft is necessary to the development of these tools only serves to benefit them, because it makes people think that it's either AI built on theft or no generative algorithms at all, but that simply isn't the case.

Generative algorithms are like, really simple when you break it down. All you need is a lot of inputs, a lot of trainers, a lot of processing power, and a lot of time. All of those cost money, but if you steal, you can decrease the time by increasing the inputs, saving money on both time and inputs. And the entire reason this has only now become a problem is that only in the past few years can your average slightly-well off schmuck get access to the processing power needed.

Back in the early 2010s only big corporations could develop these kinds of algorithms, and they were costly to maintain, not open to the public, and didn't use unlicensed materials for fear of lawsuits. Now though, relatively young startups have shown the tech world that if you throw enough money at PR and convince the public that your Plagiarism Machine is synonymous with the entire concept of machine learning, then you can completely control the press you get.

This just. Pisses me off so much. I literally studied AI as a minor in college, before ChatGPT and DALL-E hit the scene and changed everything. Machine Learning, or "AI", can do so much good -- it's helped scientists identify diseases, detect cancer, sort the lineages of living and extinct animals, it's detected serious coding errors and automated menial tasks for artists and engineers alike -- but it's become so associated with evil. And that just breaks my heart.