deks

EE Eng. autonomous drone swarms and robot learning

[deks @deksprime]

@AnthropicAI 🥱

[deks @deksprime]

@cursor_ai Composer 2.5 marketing goes hard

[deks @deksprime]

@willccbb @bllchmbrs @tautologer It's really cute that everyone believes that they're doing it because of safety. It's their futile attempt to prevent competition and that's it.

[deks @deksprime]

@Teknium So that people lose time with it, thinking it will actually do something, but it won't, I doubt that it ever will. And in return, more resources can be allocated to actual models...

[deks @deksprime]

now is the best time to use Opus 4.8 while everyone's trying to get Fable to respond

[deks @deksprime]

Breaking news: You're not a lion. Every single "prey" of yours can outsprint you. Yes, sprinting is the best form of running, and yes, long-distance running lowers T, but the "hunter" thesis is outright wrong. Humans depend on their intelligence and collaboration to hunt, not sprinting.

[deks @deksprime]

@chesterzelaya no optical flow?

[deks @deksprime]

@DominiqueCAPaul Bro where's your kill switch 😆

[deks @deksprime]

@MercatJean this is exceptional, great job!

[deks @deksprime]

@consequence definitely doesn't look like the real thing

[deks @deksprime]

We are so fucking back

[Davide Scaramuzza @davsca1]

We are excited to share our #ICRA2026 paper "Dream to Fly: Model-Based Reinforcement Learning for Vision-Based Drone Flight"!
Paper: https://t.co/lYq77VhVvn
Video: https://t.co/Fhm5RA8aGE

Can we use Model-Based #ReinforcementLearning (MBRL) to fly a drone from pixels to commands? In this work, we train quadrotor navigation policies from scratch using #WorldModels, mapping raw onboard camera pixels directly to control commands, much like a human pilot!

While model-free methods like PPO are sample-inefficient and struggle in this setting, we leverage #MBRL to train visuomotor policies capable of agile flight through a racetrack using only raw pixel observations, no explicit state estimation needed.

A key finding: because our policies are trained end-to-end directly from pixels, we no longer need the perception-aware reward term used in previous methods. Instead, this behavior emerges naturally! The policies learn to guide the camera toward feature-rich areas of the observation space on their own.

Kudos to @roaguiangel @ashwinshx @isgeles @EliJalbout

Reference:
"Dream to Fly: Model-Based Reinforcement Learning for Vision-Based Drone Flight"
Angel Romero*, Ashwin Shenai*, Ismail Geles, Elie Aljalbout, Davide Scaramuzza IEEE International Conference on Robotics and Automation (ICRA), Vienna, 2026.

@ERC_Research @AUTOASSESS_EU @uzh_ifi @UZH_en @UZH_Science @Prophesee_ai @SynSenseNeuro @UZHspacehub @swissrobotics @nccrrobotics

[deks @deksprime]

@reach_vb wtf are we doing here anymore man

[deks @deksprime]

@brysonkjones This is great work! Are you planning to share the weights any time soon? I'd really love to add the results of the LIBERO version to roboticinference[.]com/leaderboard. And also the base model for usage and fine-tuning.

[ben hylak @benhylak]

the year is 2027.

garry tan has just crossed 1b lines of code per day. water to 3 rural californian towns were diverted in order to cool his locally ran LLMs.

riots erupt, and protesters demand answers to one, single question:

"what is he building?"

[deks @deksprime]

37K LOC

> .md files

[Garry Tan @garrytan]

Absolutely insane week for agentic engineering

37K LOC per day across 5 projects

Still speeding up

[deks @deksprime]

@noahzweben nice RL tasks

[deks @deksprime]

@barkmeta That ship has sailed bro, long gone. There are already 500M+ smartwatch users worldwide. Users that actually PAY money to get tracked. You can't scare them with this cute little cow allegory.

[deks @deksprime]

@irvinxyz yeah, not yet.

[deks @deksprime]

@Sajeel_Purewal "Did their entire feasibility + system design in under 5 minutes"

it shows.