New SAGE planner enhances latent world model long-horizon planning

Researchers have developed SAGE, a new planning paradigm for latent world models that improves long-horizon planning by generating subgoal-conditioned actions. This method uses a goal-conditioned generator to predict reachable latent subgoals of varying durations, which then guide the generation of candidate action sequences. Experiments on PushT and OGBench Cube demonstrate significant improvements in planning success rates, with PushT success increasing from 12.7% to 64.7% and OGBench Cube success from 26.7% to 67.3% when targeting a 150-step offset. AI

IMPACT This new planning approach could enable more effective long-horizon decision-making in AI systems.

RANK_REASON The cluster contains an academic paper detailing a new method for AI planning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Latent world models
• SAGE

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →