Senior Research Engineer @UCL @ucl_arc, driving innovation in #physicalAI for #Healthcare #SurgTech #MedTech & #ClinicalTranslation Innovating @air4children
London/Oxford, England
Joined May 2015
- Tweets8.1K
- Following2.5K
- Followers1.3K
- Likes26.1K
Pinned Tweet
๐ข Exciting news! The abstract submission deadline for our full-day workshop, "Healing Through Collaboration: Open-Source Software in Surgical, Biomedical and AI Technologies", has been extended to 12 June 2025, 12:00 PM (UTC+0).
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
20 Open-Source Humanoid Robots You Can Build Yourself | From $50 to $900
What if you could build your own humanoid robot without spending thousands of dollars?
In this video, I explore 20 open-source and DIY humanoid robotics projects, ranked from 1 to 20. The list covers a wide range of budgets, from around $50 to $900 for an upper-body build.
Some projects provide official 3D models, CAD files or assembled STEP models. Others do not have a complete, ready-to-use 3D model matching the robot shown in the video.
For some entries, I arranged available components or created my own visual representations. The 3D models for projects 2, 8, 10 and 6 require verification. Some visuals may be reconstructions or assemblies rather than official models of the original robots. Please check the project documentation before treating any visual as an exact representation of the hardware.
The 20 projects featured
1. RoboPrime | Around $60โ100
A low-cost robotics project with an open-source design.
2. Open-source DIY humanoid builds
A general category covering custom humanoid builds assembled from different hardware components. The humanoid shown is a generic 3D model created by me, not an official robot model.
3. Qmini | Price unconfirmed
A compact robot project from Unitree Robotics.
4. Zowi | Around โฌ100
A small educational robot associated with BQ.
5. Otto Humanoid / Otto DIY | From around $50
A DIY robotics platform. The final cost depends on the selected components and configuration.
6. Kayra | Price unconfirmed
The visual combines official leg designs with additional components positioned by me.
7. PLEN2 | Around $499 for the kit
A compact humanoid robot with published CAD files and walking-related software.
8. Nova Desktop Humanoid | Around $150
A desktop humanoid project by Daniel Hany. I positioned the available head and arm components myself in the 3D scene.
9. Bambot | Around $300
A low-cost robotics project with open-source design resources.
10. Rapiro | Around $390
A small humanoid robot designed by Shota Ishiwatari. The 3D reconstruction shown in the video requires verification against the original model.
11. Valera | Around $400
An open-source humanoid robotics project.
12. Reachy Mini Lite | Around $399
A compact robotics platform developed by Pollen Robotics in collaboration with Hugging Face.
13. Open Duck Mini | Under $400
An open-source bipedal robot project for robotics enthusiasts and developers.
14. MicroDuck | Around $399
A Pollen Robotics project with reinforcement-learning resources.
15. ROBOTIS MINI | Around $540
A compact humanoid platform associated with ROBOTIS components.
16. Halley | Around $500โ800
An open-source robot project with 3D design resources.
17. Microban | Around $550โ600
A robotics project associated with the Rhoban team at the University of Bordeaux.
18. Endeavor II | Around $600
A humanoid project with an assembled STEP model.
19. XLeRobot | Around $660
An open-source robotics project with accessible hardware designs.
20. InMoov | Around $900 for the upper body
Created by Gaรซl Langevin, InMoov is a well-known 3D-printed humanoid robotics project.
What I noticed while preparing this video
The projects vary in price and in the amount of design information available. Some provide official 3D models and detailed mechanical files, while others need additional work to create a visual representation.
The models shown for projects 2, 6, 8 and 10 need to be distinguished from verified official 3D assets. Check the original repositories and documentation for confirmation.
This completely changes how robots can learn in simulation.
NVIDIA just launched SimReady Central with 3,500+ simulation ready OpenUSD assets free and ungated on Hugging Face.
โ Robot arms grippers hospital equipment warehouse gear kitchens.
Lightwheel is the largest contributor to the library. Its assets are physics validated with measured mass and inertia tuned collision geometry and joint limits that hold under contact.
These are more than pretty meshes.
โ A robot in simulation learns from how objects collide how heavy they are what moves and how far.
If an object looks right but behaves wrong it teaches the policy the wrong lesson.
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Egocentric devices are becoming everyday wearables. If any of them can act as a full-body mocap system, everyday life becomes a scalable source of 3D human motion data for embodied AI and robot learning.
Our OmniEgoCap is a step in that direction. Great work by @kyungwon__cho and @nagooon317!
#NeurIPS2026
Turn any egocentric device into a full-body 3D mocap system!
We present "OmniEgoCap" at #NeurIPS2026.
Try the interactive 3D demo ๐ kyungwoncho.github.io/OmniEgโฆ
โข ๐๐ป๐ฝ๐๐: Head trajectory and whichever hands the device sees
โข ๐ข๐๐๐ฝ๐๐: Full-body motion with a single body shape
โข ๐๐ฒ๐๐ถ๐ฐ๐ฒ-๐๐ด๐ป๐ผ๐๐๐ถ๐ฐ: One model for Aria, GoPro, Quest 3 and iPhone, no retraining
โข ๐๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ: OmniEgoDB, the first mocap benchmark across multiple consumer egocentric devices
๐ป Code & Checkpoints: github.com/KyungWonCho/OmniEโฆ
๐ฆ OmniEgoDB: huggingface.co/datasets/cscaโฆ
๐ Paper: arxiv.org/abs/2512.19283
Joint work with Jeonghyeon Na (@nagooon317) and Hanbyul Joo (@jhugestar) at Seoul National University.
Huge thanks to my collaborators!
Details in the thread ๐
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Collecting data for dexterous hands is hard. Teleop needs a full setup. Bringing UMI-like devices to 20-dof hands is challenging.
Introducing DexPort ๐๏ธ: a hand-held interface w/ a real robot hand. No embodiment gap, recorded joint state AND actions, easy to use.
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
My first open model lecture, covering all the many pieces that build the ecosystem, and the open questions/debates determining where it goes next.
How does distillation fit in?
How do we manage risks?
How do we see access to models changing in the future?
I'm excited to revisit these questions about every 6 months, to give a state of the union on the field.
Chapters
00:00 Introduction
01:36 What is an open model?
08:47 Why should we care about open models?
18:52 How competitive are the best open models today?
25:28 How are open models built? The role of distillation
34:52 Are people actually using open models?
41:47 What comes next with open models?
47:32 Conclusions
What questions did I not answer that you have?
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
๐ After one year of robot pretraining, we learned one thing the hard way: data is everything.
Today we're excited to release FOMI, a Force-feedback Manipulation Interface for teaching contact-rich manipulation!
Inspired by @SongShuran's UMI, FOMI keeps handheld, robot-free data collection and adds what operators have been missing: you can feel what the gripper touches โ No force sensors, no motors.
Two key results:
๐ 29% more successful demos than UMI in the same time, and 18% more than GELLO in teleoperation
๐ ฯ0.5 fine-tuned on about an hour of FOMI data per task reaches 78.4% average success on four contact-rich tasks
Let's check out our demos ๐
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
A paper in Nature describes a multi-agent framework that coordinates data science, medical reasoning and health coaching over longitudinal wearable and clinical data to provide holistic personal health support. go.nature.com/4k8aahI
ALT This is figure 1, which gives an overview of the personal health agent
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Weโve built a customizable world model that turns just a few photos into an interactive virtual environment.
Let's transform real-world scenes directly into virtual spaces ready for exploration and manipulation!
Project: mundo-world-model.github.io
Blog: huggingface.co/blog/Luo-Yihaโฆ
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
One world, one policy.
MorphMimic shares a morphology-conditioned world model and policy across #Unitree G1, #AgiBot X2 Ultra, and #Booster T2โfrom the same dance in MuJoCo to a live browser demo and Unitree G1 hardware, with more to come. Code and report at lzyang2000.github.io/morphmiโฆ
Huge thanks to the SuperMimic team for the inspiration!
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Chinese eye hospital reports the country's first retinal-pathway visual implant in a blind patient neurotechguide.org/news/chinโฆ retinal implant, #BCI #blindness
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published
@Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index.
Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3.
Key results:
โค Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%.
โค Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%.
โค Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower.
Additional model details:
โค Context window: 1M tokens.
โค Size: 2.8T total, 104B active parameters
โค Input modalities: Text and image
โค Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch
โค Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates.
Congratulations to the team at @Kimi_Moonshot on the release!
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Reset incoming!
Teams are already hitting their weekly limits on Open Math OS, so weโre resetting usage for everyone.
Keep doing great math!
Apply for Open Math OS:
openmathmodel.org/
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
I love it when technology feels like magic.
Giving SAM to an agent feels like magic to me.
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Physical Intelligence's other co-founder put her entire Stanford course online: 18 lectures, this spring's slides and the homework, free.
"CS224R: Deep Reinforcement Learning," Chelsea Finn (@chelseabfinn), Stanford.
Imitation learning, policy gradients, actor-critic, Q-learning, offline RL, reward learning, RL for LLM reasoning, model-based RL, multi-task and meta RL, exploration and RL for robots - 18 lectures plus a tutorial, about 20 hours, on Stanford Online's YouTube.
Lecture 16 is RL for real robots. Lecture 17 is a guest lecture from Ashish Kumar, who led AI for Tesla Optimus.
1X is hiring an AI Researcher for robot learning right now, at $200Kโ$300K. this course is the research that job title runs on.
Lectures: youtube.com/playlist?list=PLโฆ
Slides + homework: cs224r.stanford.edu
P.S. yesterday it was Sergey Levine's Berkeley course. between the two, both Physical Intelligence co-founders are now teaching you for free.
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
OlmoEarth v1.3 is out. ๐ State-of-the-art embeddings from the same foundation model family, precomputed for every 10 m pixel on Earth.
Weโre excited to give you more ways to work with OlmoEarth, including a new browser tool to explore and customize maps.
Full rundown in ๐งต
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
The next frontier is AI that improves itself, and we believe everyone should know how to build it.
This semester, Ao @ao_qu18465 and I are teaching 'How to AI an AI for Almost Anything' @MIT @medialab -- a new course on self-evolving AI systems that recursively improve from experience, feedback, memory, and exploration over long interaction horizons.
The key idea: intelligence shouldn't stop improving when training ends.
We cover four levels of evolution:
(1) context evolution through prompt optimization and context management,
(2) harness evolution through memory, tools & skills, workflow optimization, and multi-agent systems,
(3) model evolution through RL, distillation, test-time training, self-play, and continual learning,
(4) task evolution through generating increasingly complex synthetic tasks, simulations, evaluations, and environments.
Slides, readings, and videos available: mit-mi.github.io/how2ai-courโฆ
We thank @Kimi_Moonshot and @PrimeIntellect for generously sponsoring API credits!
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Open Math Harness is open source: the runtime of Open Math OS, infrastructure for Math 2.0.
Built for everyday math research: agents on open models with math and scientific computing tools, parallel exploration in Ultramath, autoformalization in Lean.
github.com/openmathmodel/opeโฆ
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Ubuntu 26.10 #StonkingStingray is now available.
Read our announcement for more details: ubuntu.com/blog/canonical-reโฆ
#Ubuntu #Linux #OpenSource
Miguel Xochicale ๐ง๐ฝโ๐ฌ๐คใฐ๏ธ retweeted
Security updates are available for CVE-2026-109618 and CVE-2026-109619 (php2shell), two critical vulnerabilities affecting PHP that could allow unauthenticated remote code execution.
Canonical has released fixes for all affected supported Ubuntu releases. Users should update affected systems as soon as possible.
Learn more and review update instructions: ubuntu.com/blog/php2shell-vuโฆ
#Ubuntu #Security #CVE
