The paper proposes Interpretable Continuous Control Trees (ICCTs), a tree-based model that can be optimized using modern, gradient-based, reinforcement learning approaches to produce high-performing, interpretable policies. The ICCTs show potential in learning policies that outperform baselines by up to 33% in autonomous driving scenarios while achieving a 300x-600x reduction in the number of parameters against deep learning baselines. The ICCTs are found to be easier to simulate, quicker to validate, and more interpretable than neural networks, making them ideal for usage in safety-critical and legally-regulated domains.
Publication date: 16 Nov 2023
Project Page: https://arxiv.org/abs/2311.10041v1
Paper: https://arxiv.org/pdf/2311.10041
