Kickoff day for the Summer Robotics Challenge: three months to build something with embodied AI. Robotics Nation is running it.
They opened with a question: what would you do if you had your own embodied AI assistant for the whole summer? Simple to ask. Harder to answer.
"Hot bot summer" works on a few levels for me. It's the meme. It's literally summer. And at some point the Jetson on my bench is going to run hot too.
This isn't a weekend hackathon or a stage demo that falls apart the second you unplug something. It's twelve weeks to break things, rebuild them, and land on an MVP that a real person might pay for.
the deal
Each team gets an XLerobot kit: open-source hardware, OAK-D depth camera, Meta Quest 3S, Jetson Orin Nano. About €1,500 of parts.
There's also a €7,000 prize pool. Demo day is September 13, and the top teams get to show at Robotgames in late October.
Teams can be one to five people. I'm going solo. Not because I couldn't find anyone, but because I want my hands on every part of it: the wiring, the models, the bits that break at 1am. If I'm going to relearn this from scratch, I don't want to be able to hide behind someone who already gets it.
why this, why now
I haven't competed in robotics since RoboCup Junior 2017. Three national finals in India, a trip to Japan to represent the country. Before that it was LEGO Mindstorms in 2015, then a ROS 2 project in 2022. Everything since has been software: startups, Slush, things scaling to tens of thousands of users.
So I'm rusty, and I've decided that's fine. Coming back now, the field barely looks the same. The robot used to be the hard part. You wrote the logic by hand and the machine did exactly what you told it, mistakes included. Now a lot of the intelligence lives in the models: big VLA systems that look at a scene and decide what to do, more like steering something that already has opinions than programming a state machine. I learned on color sensors and open-loop turns, so this is a different way of thinking about control. I'm not going to out-read the people who've lived in this for years, so the plan is to learn by building: read enough of the key work to know what's possible, then get my hands on the bot and find out where my mental model is wrong. Fresh eyes might be the only edge I've got.
But I've never really left building. LEGO on the floor in New Delhi. A research assistant gig at Aalto on a robotic arm. Same itch as when I was a kid stacking blocks to see if they'd hold.
Embodied AI feels like the obvious place to pick that back up. Code in a physical body has to deal with actual space and actual objects. It fails in ways a browser tab never will, which sounds more fun than another dashboard at this point.
the plan
I'm going to post here every week for the next twelve. What I tried, what broke, what I'd redo. The stuff that doesn't end up in the demo reel.
A few dates:
- 13 June: kickoff and hardware handoff (today)
- 13 September: demo day in Otaniemi
- 30 Oct to 1 Nov: Robotgames for top teams
I still haven't picked the MVP. Grab something from the fridge? Fold laundry? Water plants while you're away? It needs to be boring enough that someone would pay for it and weird enough that building it actually teaches me something.
week 0
Today was mostly logistics: meet the other teams, poke at the XLerobot, try to figure out what "three months" feels like when the hardware is actually in your hands.
Small group, and everyone was keen in that slightly nervous way you get on day one, before anyone has anything to show. No bravado yet. Just a room of people who'd signed up to spend their summer on this.
The hardware made it real. Picking up the XLerobot, plugging in the Jetson, looking at the OAK-D and the Quest sitting there waiting. I'm genuinely excited to break it. Not carefully. I want to push it until something gives, because that's usually where the interesting problems are.
No breakthroughs. Just week 0. But I haven't started something in a while without knowing how I'll finish it, and I kind of missed that feeling.
See you next week.