GRASP: Making Long-Horizon AI Planning Actually Work with World Models
Berkeley researchers introduce GRASP, a breakthrough approach that solves the notorious challenges of long-horizon planning with AI world models through clever gradient optimization tricks.