Stage 14 · Advanced Robotics / Reinforcement Learning · Lesson 5 · 14–20 min

Reward Shaping for Manipulation

Sparse pick success (+1 at the end) is honest but hard. Dense/shaped rewards teach faster — and can be hacked. This lesson sets up later Parts.

• All Stage 14 lessons
• Previous: Episodes, Exploration vs Exploitation
• Next: Why Tabular Q Breaks for Robots