Peter Luo retweeted
Introducing Kinex, the harness that connects coding agents to any physical robot, unlocking scalable learning in the physical world.
Tell Kinex about your robot; it then carries out tasks and learns from every run. We demonstrate Kinex across diverse real-world applications in a zero-shot setting, from bimanual manipulation to quadruped robots.
🧵1/5
kinex, harness for intelligent physical interaction! #kinex
I’ve been thinking a lot about how AI can learn from interacting with the physical world. We learn by trying things, seeing what happens, and adjusting. The same feedback loop is essential if we want AI to interact with the world as naturally as we do.
With Kinex, we’re building that loop between AI and robots, turning physical interaction into opportunities for learning and improvement. We’ve demonstrated zero-shot task execution on real hardware, from bimanual manipulation, mobile manipulation, to tasks with quadruped robots. No demonstration and no training.
I’m really excited to share this work, and grateful to the team for making it happen! Give Kinex a try on your robots. We’d love to hear what works, what doesn’t, and what you’d like to do with it.
1
16
Peter Luo retweeted
Excited to share RLE-Bench, a joint effort from our group and Bo’s group, @daibond_alpha .
Robotics engineering is more than writing controllers. Can AI agents learn policies, debug systems, reason about physics, build tools—and even design robot hardware?
We built 48 tasks to find out and are welcoming everyone to contribute more tasks.
Great work by the team @haitongma5426, @chenxiaogao @rqiang_rushi @daibond_alpha !
Glad to be reported by Harvard SEAS News!
RLE-Bench 2.0 is actively call for tasks and proposals:
rle-bench.github.io/contribu…
Submit and understand how coding agents work in your daily workflow!
GPT-6 Astra is impressive, but agentic robotics ≠ just control.
Physical agents should control robots, learn new skills, and design their own hardware.
So we introduce RLE-Bench: 48 everyday robotics engineering tasks spanning closed-loop control, policy learning, perception, and mechanical design.
Following up on the questions about manufacturability: yes, we checked. The design tasks use either aluminum profiles or 3D printing, so they do not require specialized manufacturing processes. Assembly video shown below ↓
The gap is not just manufacturability — current coding agents still have weak physical reasoning around stability and dynamics.
Recent RoboDojo results by @MarioChan2002 have also highlighted a lack of “physical commonsense.” What we see in RLE-Bench is that this problem goes well beyond control.
Peter Luo retweeted
GPT-6 Astra can solve some seriously tricky spatial constraint tasks: from unlocking interlocked parts to threading a rope through three rings.
Try it yourself now 👇
qinengwang-aiden.github.io/d…
The ChatGPT moment for Physical AI may be closer than we think.
Check out RLE-Bench, a new benchmark for evaluating AI agents across the entire robotics stack, from perception and control to policy learning, deployment, and mechanical design.
rle-bench.github.io/
GPT-6 Astra is impressive, but agentic robotics ≠ just control.
Physical agents should control robots, learn new skills, and design their own hardware.
So we introduce RLE-Bench: 48 everyday robotics engineering tasks spanning closed-loop control, policy learning, perception, and mechanical design.
1
3
331
AI Security Weekly: The Week Agents Got a Security Bill of Rights #AISecurity open.substack.com/pub/peterl…
1
10
Check out my latest ai security newsletter Mar 8-15, 2026! $395M flooded into AI agent security this week — because #OpenClaw just proved every autonomous #agent in your enterprise is a ticking liability. open.substack.com/pub/peterl…
1
23
83% of orgs deploying AI agents in 2026. Only 29% can secure them.
This week: $76M raised for agent security, NIST launched its AI agent standards initiative, and shadow AI + agent identity just became your biggest unsolved problem peterluo.substack.com/publis…
1
14
BREAKING: 30+ countries warn AI systems could enable hacking/bioattacks; First AI agent breach predicted MID-2026; AI identities flooding IAM systems
Traditional security is dead.
AI-native defense or bust.
#Cybersecurity #AI open.substack.com/pub/peterl…
16
AI Security Weekly: Jan 19-25: Mandiant founder Kevin Mandia launches Armadin ($24M seed, $600M+ valuation) predicting 'offense will be all-AI in 2 years.' HiddenLayer wins $151B DoD contract, Chrome extensions leak 900K users' AI chats. open.substack.com/pub/peterl… #AI #Security
1
85
Jan 12-18: Delinea acquires StrongDM for AI-era identity security, Aikido hits $1B unicorn, while n8n suffers coordinated attack. #AI #Security open.substack.com/pub/peterl…
16
OWASP's Top 10 for Agentic Applications—a good start, but the agentic industry is still evolving, expecting more to come in 2026. open.substack.com/pub/peterl…
19
Cybersecurity + AI Weekly Newsletter 12/2025 ServiceNow's $7.75B Armis acquisition + LangGrinch Vulnerability open.substack.com/pub/peterl…
1
43
A wild week in AI security: ICLR reviewer identities leaked, Google spotted self-modifying AI malware , and Clover raised $36M to catch vulnerabilities before code is even written. #cybersecurity #AI open.substack.com/pub/peterl…
1
389








