Situation-aware Autonomous Driving Decision Making with Cooperative Perception on Demand

TL;DR

Problem: urban autonomous driving requires robust situation awareness; cooperative perception can extend perception but is restricted by wireless bandwidth. Method: propose Cooperative Perception on Demand (CPoD) integrated into a POMDP framework and solved online with DESPOT; extended sensing when active is defined as $\mathbf{Z}^{[e]}(a_{\text{CPoD}})=Z^{[e]}\cup Z^{[iov]}\otimes Q^{[iov,e]}$, and the objective is $\mathbb{E}\left[\sum_{t=0}^\infty \gamma^{t} \mathcal{L}(a_t,s_t)\right]$. Contributions: formalization of CPoD within POMDP, belief tracking for obstacle motion intentions, a two-component reward $R(s,a)=R(s,a_{ACC})+R(s,a_{CPoD})$ with $R(s,a_{CPoD})=R_{comm}+R_{intention}+R_{TTC}$, and online evaluation in urban scenarios showing improved safety and reduced communication. Significance: enables safer, more efficient urban driving under occlusion and limited communications by activating cooperative perception only when needed.

Abstract

This paper investigates the impact of cooperative perception on autonomous driving decision making on urban roads. The extended perception range contributed by the cooperative perception can be properly leveraged to address the implicit dependencies within the vehicles, thereby the vehicle decision making performance can be improved. Meanwhile, we acknowledge the inherent limitation of wireless communication and propose a Cooperative Perception on Demand (CPoD) strategy, where the cooperative perception will only be activated when the extended perception range is necessary for proper situation-awareness. The situation-aware decision making with CPoD is modeled as a Partially Observable Markov Decision Process (POMDP) and solved in an online manner. The evaluation results demonstrate that the proposed approach can function safely and efficiently for autonomous driving on urban roads.

Figures (7)

• Figure 1: Illustrative example of CPoD: The blue obstacle vehicle is blind the red ego vehicle. Given the shared observation from the IOV, ego vehicle can conduct a more comprehensive situation reasoning.
• Figure 2: System framework for situation-aware decision making with CPoD.
• Figure 3: Evaluation environment and scenario: (a) shows the evaluation environment, which is a typical urban road within the campus of National University of Singapore. (b) depicts the evaluation scenario, where the red points represent the ego vehicle's laser scan and the IOV's laser scan is shown by the green points.
• Figure 4: Navigation process of situation-aware decision making with CPoD at T-junction. The actions are highlighted with the colorful text. For acceleration decision: [Red: $\textsc{Decelerate}$, Green: $\textsc{Accelerate}$, Blue: $\textsc{Constant}$]. For CPoD decision: [Red: $\textsc{Deactive}$, Green: $\textsc{Active}$] The motion intention is represented as the cubic marker: [Sky Blue: $Stopping$, Brown: $Hesitating$, Yellow: $Normal$, Purple: $Aggressive$], where the cubic height is proportional to the corresponding belief value.
• Figure 5: Situation evolution analysis: (a) shows the obstacle clearance and (b) represents the acceleration command along with the speeds of both IOV and ego vehicle. (c) depicts the CPoD decision with the IOV's Normal belief, and (d) illustrates the obstacle vehicle's motion intention belief.