Robot Learning
Last updated: Sep 4, 2023
Overview
Robot learning is a multidisciplinary field at the intersection of machine learning and robotics, focused on enabling robots to acquire know-how and skills through experience. The goal is for robots to adapt and improve their performance in various tasks autonomously. This field encompasses many learning paradigms: supervised, unsupervised, reinforcement, imitation, self-supervised, and meta-learning, among others. Notable goals for robot learning include: perceive and understand the environment, plan actions, manipulate objects, and interact with humans and other robots. While robot learning admits many potential applications (e.g., navigation, manipulation, interaction), there are many obstacles to deployment; for example, data requirements, safety concerns, lack of explainability and interpretability, data quality and bias.
Fundamentals
In progress.
Papers of Interest
The links below are not full citations but are listed for brevity (date, title, first author). This does not intend to discredit any co-authors. For full attribution, please visit the links.
- 2021, A review of robot learning for manipulation: challenges, representations, and algorithms, Kroemer et al.
- 2023, RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control, Brohan et al.
Resources
- Robot Learning Foundation (CoRL Governing Body)
- UC Berkeley – Robot Learning Lab
- Microsoft Research – Robot Learning Group
- CS391R: Robot Learning
- UT Austin – Robot Perception and Learning Lab