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.
13
33
224
169,717
Nice work Chenxiao! Try kinex ⬇️
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
2
159
rlebench 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
3
11
25
71,767
Two updates on RLE-Bench! 🤖
📈 New results for GPT-6.1-Sol, Opus 5.5, Fable 5.1, GPT-6-Luna, and Grok 4.6! Opus 5.5 shows substantial gains in interactive control—even outperforming Fable 5.1. Exciting to see frontier models making progress on robotics tasks!
📑 Our technical report is out! We dive into the design principles behind RLE-Bench, break down each task, and share observations of how agents tackle them. Plenty of interesting agent behaviors to explore!
Check out the latest results and read the report 👇
1
5
494
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!
1
1
165
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.
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
1
10
2,784
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.
13
33
224
169,717
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.
13
33
224
169,717
(5/5) Huge thanks to our core contributors @haitongma5426 @chenxiaogao @rqiang_rushi
and our advisors @daibond_alpha
and @Na_Lina_Li!
Our RLE-Bench 2.0 is calling for contributions, check out here rle-bench.github.io/contribu…!
4
559
