Nazakat: Compliant Morphology and Developmental Learning for Robotic Manipulation

Robotic manipulation of unknown objects remains difficult because a gripper cannot know an object's firmness, shape, or slipperiness in advance.We present Nazakat (Naz), a controlled 2 × 2 factorial simulation study crossing gripper morphology (rigid vs. cephalopod-inspired soft) with learning strategy (sparse-reward reinforcement learning (RL) vs. a six-stage developmental curriculum modeled on infant motor milestones), trained with Soft Actor-Critic and evaluated on twelve objects across three random seeds per condition.Compliance produced the largest and most consistent effect: the soft gripper improved generalization over the rigid gripper in every seed-matched comparison under both learning strategies, although soft-gripper standard RL was also NOT PEER-REVIEWED