# Comparing Human-Centric and Robot-Centric Sampling for Robot Deep Learning from Demonstrations

arXiv: https://arxiv.org/abs/1610.00850 · HTML: https://mlsanity.com/paper/1610.00850/

## Key facts

- Cited by 17 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 4 Oct 2016.

## Built on by (most cited first, 17 of 17)

- [HG-DAgger: Interactive Imitation Learning with Human Experts](https://mlsanity.com/paper/1810.02890/) (2018)
- [Imitation Learning as f-Divergence Minimization](https://mlsanity.com/paper/1905.12888/) (2019)
- [Human-in-the-Loop Imitation Learning using Remote Teleoperation](https://mlsanity.com/paper/2012.06733/) (2020)
- [Feedback in Imitation Learning: The Three Regimes of Covariate Shift](https://mlsanity.com/paper/2102.02872/) (2021)
- [Interactive Robot Learning from Verbal Correction](https://mlsanity.com/paper/2310.17555/) (2023)
- [AR2-D2:Training a Robot Without a Robot](https://mlsanity.com/paper/2306.13818/) (2023)
- [IntervenGen: Interventional Data Generation for Robust and Data-Efficient Robot Imitation Learning](https://mlsanity.com/paper/2405.01472/) (2024)
- [Generalization Through Hand-Eye Coordination: An Action Space for Learning Spatially-Invariant Visuomotor Control](https://mlsanity.com/paper/2103.00375/) (2021)
- [On-Policy Robot Imitation Learning from a Converging Supervisor](https://mlsanity.com/paper/1907.03423/) (2019)
- [Convergence of Value Aggregation for Imitation Learning](https://mlsanity.com/paper/1801.07292/) (2018)
- [ARCADE: Scalable Demonstration Collection and Generation via Augmented Reality for Imitation Learning](https://mlsanity.com/paper/2410.15994/) (2024)
- [Dynamic Regret Convergence Analysis and an Adaptive Regularization Algorithm for On-Policy Robot Imitation Learning](https://mlsanity.com/paper/1811.02184/) (2018)
- [Counterfactual Behavior Cloning: Offline Imitation Learning from Imperfect Human Demonstrations](https://mlsanity.com/paper/2505.10760/) (2025)
- [Improving Robot-Centric Learning from Demonstration via Personalized Embeddings](https://mlsanity.com/paper/2110.03134/) (2021)
- [Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning](https://mlsanity.com/paper/2412.07057/) (2024)
- [Investigating the Effects of Robot Engagement Communication on Learning from Demonstration](https://mlsanity.com/paper/2005.01020/) (2020)
- [Sample-Efficient Expert Query Control in Active Imitation Learning via Conformal Prediction](https://mlsanity.com/paper/2512.00453/) (2025)

## Abstract (arXiv)

Motivated by recent advances in Deep Learning for robot control, this paper considers two learning algorithms in terms of how they acquire demonstrations. "Human-Centric" (HC) sampling is the standard supervised learning algorithm, where a human supervisor demonstrates the task by teleoperating the robot to provide trajectories consisting of state-control pairs. "Robot-Centric" (RC) sampling is an increasingly popular alternative used in algorithms such as DAgger, where a human supervisor observes the robot executing a learned policy and provides corrective control labels for each state visited. RC sampling can be challenging for human supervisors and prone to mislabeling. RC sampling can also induce error in policy performance because it repeatedly visits areas of the state space that are harder to learn. Although policies learned with RC sampling can be superior to HC sampling for standard learning models such as linear SVMs, policies learned with HC sampling may be comparable with highly-expressive learning models such as deep learning and hyper-parametric decision trees, which have little model error. We compare HC and RC using a grid world and a physical robot singulation task, where in the latter the input is a binary image of a connected set of objects on a planar worksurface and the policy generates a motion of the gripper to separate one object from the rest. We observe in simulation that for linear SVMs, policies learned with RC outperformed those learned with HC but that with deep models this advantage disappears. We also find that with RC, the corrective control labels provided by humans can be highly inconsistent. We prove there exists a class of examples where in the limit, HC is guaranteed to converge to an optimal policy while RC may fail to converge.