A JAX-native multi-objective RL framework trains humanoid robots across six conflicting goals in minutes, delivering 21 to 270 times speedups.
document fetched 2026-10-31 02:55