Gen1 Mels
Mels has always lived in a tweener stage between creative sandbox gaming and AI robotics tooling. This is by design, to bridge two very different worlds to make the technology frontier accessible to everyone. Over the last few years we’ve learned a lot about how to grow up, without compromising our vision, thanks in large part to this community.
We’re pouring all of this good stuff into our next-generation Mels platform.
Gen2 Mels
Mels is becoming the playful robotics simulator we’ve always wanted — by getting real:
🦿 Real Physics: we’ve upgraded backend training and frontend inference with the engine used by SOTA labs, with big gains in performance and fidelity for small hits on time and costs
🥼 Real Science: we’ve exposed a lot of more of all the tooling that we use ourselves to visualize, triage, and debug our R&D, so you now have the same resources at your fingertips
🏡 Real Environments: we’re grounding environments in the real-world with true-to-life scale and increasingly photoreal rendering to narrow the gap between simulation and reality
🤖 Real Robots: we’re pulling all of these pieces together towards a grander goal, so you can train Mels to life in the real-world (more on this in a future “Sim2Real” post)
It’s not all upside, as making things more real means making things more complex. But what we get in return is a smoother learning curve (less janky physics and training bugs) with fewer creative limits (SOTA capabilities) and more rewarding outcomes (real-world robotics).
Demo Time
You know we like to show-not-tell, so we’ve put up a demo you can check out here:
Note: You can’t yet do most of the things you know and love in mels.ai. But you can play with our first Gen2 Mel in its home workshop environment with new physics and learning visualizations — and get an initial feel for how the whole is greater than the sum of its parts!
