Introducing HIJP: jointly learn a goal-conditioned inverse model and a world model, then plan by alternating inverse → predict.
No sampling. No search.
@DavidSunok47 Thanks for the shoutout!
INTACT is an important reference for our work.
Excited to keep pushing search-free planning with latent world models forward — and looking forward to your next surprise! 👀
Planning with a world model shouldn’t require thousands of rollouts.
Meet HIJP: train the inverse → predict loop. No test-time search.
86.8% overall success, beating 8 baselines. Up to 16.5× faster than LeWM (CEM).
Demo: https://t.co/jiQBF6jWz8
Paper: https://t.co/VV0TLPbeAL