Model recovers exact deceleration from game footage, but the author calls it a dead end
mike64_t · x · 2026-10-28
mike64t shares a counterintuitive demo: a model recovers a vehicle's real-time deceleration (in m/s²) purely from game footage, staying essentially correct even on snow — superhuman relative to humans who can't judge speed from visual cues. But he tempers the hype: it's still just a proof of concept, and if all you can do is recover these values, copy-cat policies remain a real pain, making this route a potential dead end. The game's raw acceleration data is itself noisy due to Euler-angle issues.
More from Fun
- Robotaxi meme: Waymo mogs Zoox as Cybercab watches, all three in one frame — kyliebytes · 2026-10-28
- Graduate celebrates degree with 'rm -rf school/' meme — dhruv2038 · 2026-10-28
- He Built a Game Boy Color Emulator in MicroPython on His Conference Badge — martinwoodward · 2026-10-28
- Mathematicians blast unreviewed AI proofs, then rush to publish them themselves — skdh · 2026-10-28
- Developer hooks GPT into a Hasselblad: focus, shutter and chat, open-sourced — rrRadium · 2026-10-28
- Cryptic insider tease sparks guesses: GPT-6.2 Sol or GPT-6.1 Astra? — CtrlAltDwayne · 2026-10-28