Reward Shaping Study for Sub-Goal Based Robotic Tasks

By Liliana Antão, Leonor Santos, João C. P. Reis, Gil Gonçalves. 2022

Robotics has been providing essential tools to improve the Industry’s flexibility and reconfiguration. Nowadays, the tasks are highly complex, making it mandatory to renew programming techniques and craft sophisticated solutions. This is where Reinforcement Learning comes in handy: it broadens the horizon for new and disruptive methodologies, incorporating tolerance to uncertainties using exploration. However, those solutions can be challenging when it comes to building the reward function. So, this work proposes a ‘‘divide and conquer’’ approach, which transforms a long and complex task into smaller and easier achieved goals that boosts training efficiency and performance. This study tested different combinations of parameters and proposes a set of guidelines that used a Pickand-Place environment. The observed performance shows higher success rates and faster convergence for two and three sub-goals when comparing the sub-goal approach versus a sequential and traditional reward.

Subfields: Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

Citations in the corpus

Evaluating the Social Impact of Generative AI Systems in Systems and Society (2023, background): “[14] shows how technological advancements have historically increased earning inequality between education, sex, race, and age groups.”
Image as a Foreign Language: BEiT Pretraining for All Vision and Vision-Language Tasks (2022, neutral): “We compare with previous state-of-the-art models, including FD-SwinV2 [WHX22], DINO [ZLL22], Mask DINO [ZLL22], FD-CLIP [WHX22], CoCa [YWV22], OFA [WYM22], Florence [YCC21].”