An Auto-tuning Framework for Autonomous Vehicles

Abstract

Many autonomous driving motion planners generate trajectories by optimizing a reward/cost functional. Designing and tuning a high-performance reward/cost functional for Level-4 autonomous driving vehicles with exposure to different driving conditions is challenging. Traditionally, reward/cost functional tuning involves substantial human effort and time spent on both simulations and road tests. As the scenario becomes more complicated, tuning to improve the motion planner performance becomes increasingly difficult. To systematically solve this issue, we develop a data-driven auto-tuning framework based on the Apollo autonomous driving framework. The framework includes a novel rank-based conditional inverse reinforcement learning algorithm, an offline training strategy and an automatic method of collecting and labeling data. Our auto-tuning framework has the following advantages that make it suitable for tuning an autonomous driving motion planner. First, compared to that of most inverse reinforcement learning algorithms, our algorithm training is efficient and capable of being applied to different scenarios. Second, the offline training strategy offers a safe way to adjust the parameters before public road testing. Third, the expert driving data and information about the surrounding environment are collected and automatically labeled, which considerably reduces the manual effort. Finally, the motion planner tuned by the framework is examined via both simulation and public road testing and is shown to achieve good performance.

LookOut: Diverse Multi-Future Prediction and Planning for Self-Driving

“…This is achieved by reasoning about separate contingent trajectories that are safe for each of the predicted scenarios. This stands in contrast to previous work [9,26,1,40], where a single motion plan was generated by minimizing the expected cost over all future scenarios. Such approaches, for instance, might slow down prematurely to yield to a very unlikely future trajectory from another vehicle, while our contingency planner can maintain its speed as long as it is able to stop timely before reaching the conflict region.…”

“…However, the above formulation is not exploiting the fact that only one of the predicted scenarios will happen in future and is conversely optimizing for a single trajectory that is "good" in expectation. By changing the expectation in (9) to max, the planner will optimize for the worst-case scenario regardless of the likelihood of that scenario. Consequently the planner will become over-conservative, e.g.…”

MP3: A Unified Model to Map, Perceive, Predict and Plan

“…More recently, cost map-based approaches have been shown to adapt better to challenging environments, which recover a trajectory by looking for local minima on the cost map. The cost map may be parameterized as a simple linear combination of hand crafted costs [14,46], or in a general non-parametric form [55]. To bridge the gap between interpretability and expressivity, [45] proposed a model that leverages supervision to learn an interpretable nonparametric occupancy that can be directly used in motion planner, with hand-crafted subcosts.…”

Jointly Learnable Behavior and Trajectory Planning for Self-Driving Vehicles

“…Conversely, discretization is avoided in [9], by introducing a continuous non-linear optimization method where obstacles in the form of polygons are converted to quadratic constraints. c) Learned Motion Planning: Learning approaches to motion planning have mainly been studied from an imitation learning (IL) [10], [11], [12], [13], [14], [15], [16], or reinforcement learning (RL) [17], [18], [19], [20] perspective. While most IL approaches provide an end-to-end training framework to control outputs from sensory data, they suffer from compounding errors due to the sequential decision making process of self-driving.…”