Robotics
Based on classical robotics, we construct a robot form these steps as follows: Robot Mechanics, Robot Calibration, Robot Planning and Robot Control. So I created the docs from the steps and focused on the basic principles, methods, algorithms and applications in the above categories.
JadeCong(GitHub): Awesome-Robotics
Robot Mechanics
(1) Collaborative Robot
(2) Legged Robot
(3) Mechanical Hand
(4) Humanoid Robot
(5) Soft Robot
(6) Tactile Sensor
Robot Calibration
(1) Basic Principles
(2) Methods and Algorithms
(3) Applications
Robot Planning
(1) Collision Avoidance
- Vision-Based Collision Avoidance
(2) Motion Planning
- A Star
- RRT(Rapidly Exploring Random Tree)
- RRT-Connect
- Vision-Servo Realtime Planning
(3) Task Manipulation
- Hand Grasp Dexterous Manipulation
- soft robotics
- Hand Grasp Future Direction
- 更好地理解不确定性
- 更多地利用接触
- 更灵巧地设计
- 更稳定的传感器
- Hand Grasp Methods & Skills
- (1) Transfer Learning
- Domain Adaptation
解决训练集和测试集数据分布不同的问题,从而能够实现在源域和目标域之间的自适应,进而实现少样本学习(Few/One-Shot Learning)。
- Domain Adaptation
- (1) Transfer Learning
- (2) Imitation Learning
- GAN for Imitation Learning
通过GAN网络去预测专家数据的分布,从而最小化真实数据分布和预测分布之间的差异,进而完成对agent的学习。
- GAN for Imitation Learning
- (3) Meta Learning
- Model Agnostic Meta Learning Framework
通过MAML框架能够避免强化学习的弊端,从而能够实现少样本的迁移学习,从而快速完成目标任务的学习。
- Model Agnostic Meta Learning Framework
- (4) Reinforcement Learning
- Policy Gradients
通过Policy Gradients的策略学习,从而实现对连续动作的预测,进而完成相关的目标任务。
- DDPG
通过更深层次的网络学习,从而加速学习过程,并且学到知识。
- A3C
通过分布式并行训练,从而加速网络训练过程。
- Policy Gradients
- (5) Deep Learning
- GAN
通过对抗生成网络GAN生成更多的异构分布的训练集数据,从而能够减轻收集真实数据大量的繁琐工作,同时能够到达用于测试的效果。
- LSTM
通过LSTM时间序列记忆,从而能够实现对过往的经验进行学习,并且能够理解上下文之间的联系。
- CNN
通过CNN对未知的数据进行估计,从而减轻对标签数据的依赖。
- GAN
Robot Control
(1) Position Control
- Trajectory Tracking
(2) Velocity Control
- PD Control
- PID Control
- 梯形曲线
- S形曲线
- 多项式曲线(3/5次)
(3) Force Control
- Active Control
- Impedance Control
- Dynamic-Model Control
- Passive Control
- Structure Setting
(4) Hybrid Position-Force Control
- Active Control
- Configure Space Control
- Passive Control
(5) Compliance Control
Continue reading AIRobotics