RoboDepot🤖

Your ultimate nexus point for robotics showcases and research papers, small/DIY robots and other future tech. Follow to see the robotics revolution!🔔

[Figure @Figure_robot]

Introducing Helix 02

It's our most powerful model to date - it's using the whole body to do dishes end-to-end and it's fully autonomous

[1X @1x_tech]

NEO’s Starting to Learn on Its Own

[Unitree @UnitreeRobotics]

Unitree Debuts the World’s First Humanoid Robot “App Store”🥰
Unitree welcomes users and developers worldwide to co-develop and share together. 🌹🌹
Exceptional developers will receive rewards.🌹🌹

[CyberRobo @CyberRobooo]

🤯Just saw the HuMI project, it's amazing

Ordinary people can simply carry a bag and demonstrate how to teach a humanoid robot: walking over to wipe a table, kneeling down to propose, drawing a sword, throwing toys… the whole body is coordinated like a human.

Data acquisition efficiency is 3 times higher than traditional teleoperation, and it can generalize to 70% success rate in unfamiliar rooms.

HuMI (SJTU & Tsinghua) changes the game: The cost of data collection for fully controlling a humanoid robot may even be as low as a VR headset with tracking and a mobile phone in the future.

[Nicholas Pfaff @NicholasEPfaff]

Meet SceneSmith: An agentic system that generates entire simulation-ready environments from a single text prompt.

VLM agents collaborate to build scenes with dozens of objects per room, articulated furniture, and full physics properties.

We believe environment generation is no longer the bottleneck for scalable robot training and evaluation in simulation.

Website: https://t.co/UZklSkJe9V

👇🧵(1/8)

[Humanoids daily @humanoidsdaily]

The Beijing Humanoid Robot Innovation Center has unveiled the TienKung 3.0. Equipped with 43 degrees of freedom and a new "embodied intelligence" platform, the robot is designed for autonomous operation in complex environments.

Key developments include:

Hardware Capability: The platform features high-torque joints capable of navigating 1-meter obstacles and performing precision tasks.

Open Infrastructure: X-Humanoid is open-sourcing the robot’s body design and vision-language models to reduce development barriers for the industry.

Industrial Focus: The system supports mainstream protocols like ROS2 and MQTT to facilitate integration into factory workflows.

[ARX @ARXrobotics]

AC 2
Anyone can contribute data at home

Ego+AC one+Pi*0.6

👀Uses only head cameras to complete the task.

💪The task is learned from human videos.

📱Data collection requires only a mobile phone.

📸Works across different cameras.

🦾Errors are corrected through human demonstrations.

✅The training process is simple and efficient.

🤗High task accuracy is achieved after training.

Thanks，
Tesla that video give us some ideas
Pi*0.6 awesome work！

[ARX @ARXrobotics]

the future

[Weave Robotics @weaverobotics]

Today we're releasing Isaac 0: our first robot for the home.

And we made a short video for our first customers to share what it’s been like having one in their home 😊

[Wenzhe Cai @WenzheC7616]

🤖Can robots achieve accurate navigation without any external localization feedback?
📸We present #LoGoPlanner, which handles perception, localization, and planning in one go!

Check our results on LeKiWi, G1, and Go2 robots.
🌐Project: https://t.co/mWIMzIfuqT

[Chenhao Li @breadli428]

Physical Intelligence @physical_int coffee demo at #NeurIPS2025.

What’s the hot take?

[RoboDepot🤖 @RoboDepot]

@sciencegirl The deepstate took hologram technology underground away from the public eye. This shouldve been common 2 decades ago. What they really have is probably completely mind blowing

[RoboDepot🤖 @RoboDepot]

I came back after a long time and there’s so many new startups and companies I hadn’t heard of that are now very popular.

[RoboDepot🤖 @RoboDepot]

I wonder sometimes why the method for traversing obstacles seems to be trip-then-recover as opposed to just planning steps over obstacles more precisely. I’m sure there’s a reason I just don’t understand what it is.

[Gigi YE @YJH_GIGIYE]

Small step? More like a turkey trot towards a huge feast! 🦃
Happy Thanksgiving to all! 🥧

[Lord Bebo @MyLordBebo]

🇨🇳 Northeastern University of China has developed a wheeled infantry robot

The robot is designed for combat, capable of assuming any position, moving on steep slopes, jumping over obstacles, and rotating 360 degrees. Targeting is automatic.

War is getting scarier

[RoboDepot🤖 @RoboDepot]

Very ideal embodiment

[Sourccey @sourccey]

Introducing Sourccey!

Your Personal Home Robot

[RoboPapers @RoboPapers]

Full episode dropping soon!

Geeking out with @ZhiSu22 on HITTER: A HumanoId Table TEnnis Robot via Hierarchical Planning and Learning https://t.co/wEjI62jCXh

Co-hosted by @micoolcho @chris_j_paxton

[RoboPapers @RoboPapers]

Full episode dropping soon!

Geeking out with @ZhiSu22 on HITTER: A HumanoId Table TEnnis Robot via Hierarchical Planning and Learning https://t.co/wEjI62jCXh

Co-hosted by @micoolcho @chris_j_paxton

[Mario Nawfal @MarioNawfal]

🚨NEO VS OPTIMUS: WHY TESLA'S VISION BEATS 1X'S PRIVACY TRAP

1X Technologies' Neo costs $20,000 and promises home autonomy.

Until it doesn't.

Then a 1X employee wearing a VR headset peers into your living room to manually control it.

That's not a robot. That's outsourced surveillance.

Tesla's Optimus should cost similar money but operates independently.

Deadlifts 150 pounds.

Lasts 300,000 to 500,000 miles on batteries.

Can assemble furniture, operate power tools, handle complex tasks.

Most importantly: no strangers in your home.

Neo's "expert mode" reveals the weakness in every consumer robot that prioritizes cheap autonomy over real capability.

If your robot can't work without human supervision, it's not a robot.

It's a puppet show with privacy costs.

Optimus learns. Improves. Scales. Neo needs babysitting.

Source: PCMag / @Tesla @elonmusk

[Wenli Xiao @_wenlixiao]

What if robots could improve themselves by learning from their own failures in the real-world?

Introducing 𝗣𝗟𝗗 (𝗣𝗿𝗼𝗯𝗲, 𝗟𝗲𝗮𝗿𝗻, 𝗗𝗶𝘀𝘁𝗶𝗹𝗹) — a recipe that enables Vision-Language-Action (VLA) models to self-improve for high-precision manipulation tasks.

PLD couples real-world residual reinforcement learning with standard supervised fine-tuning — letting robots discover, recover, and distill their own data flywheel.

Quick 🧵