Generative Models for Learning Robot Manipulation Skills from Humans

Chapter 2: Rewards-Driven Learning from Demonstrations

2.1 Inverse reinforcement learning of multiple reward functions by optimal policy transfer

2.2 Actor critics with experience replay for half-cheetah

References:

Chapter 3: Task-Parameterized Generative Models

3.1 Hidden semi-Markov model with linear quadratic tracking for segmentation and synthesis of robot manipulation tasks

3.2 Task-Parameterized hidden semi-Markov model for learning and reproduction of robot manipulation tasks

3.3 Task-Parameterized hidden semi-Markov model with linear quadratic tracking for opening a valve

References:

Chapter 4: Scalable Generative Models in Latent Space

4.1 Latent space representations with hidden semi-Markov model

4.2 Task-Parameterized hidden semi-Markov model with semi-tied parameters for pick and place with obstacle avoidance

References:

Chapter 5: Bayesian Non-Parametric Online Generative Models

5.1 Bayesian non-parametric clustering with mixture models under small variance asymptotics

5.2 Non-parametric scalable online sequence clustering simulations

5.3 Application to semi-autonomous teleoperation

References:

Chapter 6: Manipulation Assistance in Teleoperation

6.1 Semi-autonomous teleoperation framework for intention recognition and manipulation assistance

6.2 Semi-autonomous teleoperation of remotely operated vehicles over satellite communication in DexROV

References: