# Learning Task-Driven Control Policies via Information Bottlenecks

arXiv: https://arxiv.org/abs/2002.01428 · HTML: https://mlsanity.com/paper/2002.01428/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 4 Feb 2020.

## Built on by (most cited first, 13 of 13)

- [Fighting Copycat Agents in Behavioral Cloning from Observation Histories](https://mlsanity.com/paper/2010.14876/) (2020)
- [Run-time Observation Interventions Make Vision-Language-Action Models More Visually Robust](https://mlsanity.com/paper/2410.01971/) (2024)
- [Lossless Adaptation of Pretrained Vision Models For Robotic Manipulation](https://mlsanity.com/paper/2304.06600/) (2023)
- [Selective Visual Representations Improve Convergence and Generalization for Embodied AI](https://mlsanity.com/paper/2311.04193/) (2023)
- [Bayesian Imitation Learning for End-to-End Mobile Manipulation](https://mlsanity.com/paper/2202.07600/) (2022)
- [Rethinking Latent Redundancy in Behavior Cloning: An Information Bottleneck Approach for Robot Manipulation](https://mlsanity.com/paper/2502.02853/) (2025)
- [Masked Imitation Learning: Discovering Environment-Invariant Modalities in Multimodal Demonstrations](https://mlsanity.com/paper/2209.07682/) (2022)
- [Information-Bottleneck-Based Behavior Representation Learning for Multi-agent Reinforcement learning](https://mlsanity.com/paper/2109.14188/) (2021)
- [Review of Metrics to Measure the Stability, Robustness and Resilience of Reinforcement Learning](https://mlsanity.com/paper/2203.12048/) (2022)
- [Fundamental Limits for Sensor-Based Robot Control](https://mlsanity.com/paper/2202.00129/) (2022)
- [What to Ignore, What to React: Visually Robust RL Fine-Tuning of VLA Models](https://mlsanity.com/paper/2605.13105/) (2026)
- [Goals and the Structure of Experience](https://mlsanity.com/paper/2508.15013/) (2025)
- [Learning Manipulation-Sufficient Representations via Outcome Bottlenecks](https://mlsanity.com/paper/2609.13235/) (2026)

## Abstract (arXiv)

This paper presents a reinforcement learning approach to synthesizing task-driven control policies for robotic systems equipped with rich sensory modalities (e.g., vision or depth). Standard reinforcement learning algorithms typically produce policies that tightly couple control actions to the entirety of the system's state and rich sensor observations. As a consequence, the resulting policies can often be sensitive to changes in task-irrelevant portions of the state or observations (e.g., changing background colors). In contrast, the approach we present here learns to create a task-driven representation that is used to compute control actions. Formally, this is achieved by deriving a policy gradient-style algorithm that creates an information bottleneck between the states and the task-driven representation; this constrains actions to only depend on task-relevant information. We demonstrate our approach in a thorough set of simulation results on multiple examples including a grasping task that utilizes depth images and a ball-catching task that utilizes RGB images. Comparisons with a standard policy gradient approach demonstrate that the task-driven policies produced by our algorithm are often significantly more robust to sensor noise and task-irrelevant changes in the environment.