Preference-Augmented Distributional Reinforcement Learning for Risk-Sensitive Navigation
Preference-Augmented Distributional Reinforcement Learning for Risk-Sensitive Navigation
This research introduces a distributional reinforcement learning approach that models risk preference variations as shifts in quantile-weight distributions. By exposing agents to diverse preferences during training via preference-guided exploration and optimizing against worst-case distributions at deployment, the method aims to create conservative policies robust to multiple risk profiles without heavy reliance on external sensor signals or manual hyperparameter tuning. Experiments indicate reduced collisions and improved task success across simulated and real-world vehicle platforms.
The paper proposes a method using preference-guided exploration and worst-case distribution optimization to reduce reliance on hand-tuned hyperparameters for risk-sensitive navigation, with results demonstrated in simulated environments and vehicle platforms.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org