Using evidential occupancy grid for vehicle trajectory planning under uncertainty with tentacles

Abstract

Abstract-The uncertainty in environment perception is one of the challenges that we face in trajectory planning. For autonomous vehicle to be efficient, they need to be able to deal with this kind of uncertainty. In this work, we combine two existing frameworks: the Belief Functions to build evidential occupancy grid and clothoid tentacles for trajectory planning. First, we use evidential grids to represent the environment and the uncertainties which arise from ignorance and errors during the perception process. Secondly, we generate a set of clothoid tentacles in the egocentered reference frame related to the ego-vehicle, those tentacles represent possible local trajectories. Thirdly, we modify the evidential grid in order to take into consideration some traffic rules such as safety distance between vehicles. Then to choose the best tentacle to execute, we use reward system of a Markov Decision Process-like model to evaluate generated tentacles regarding several criteria including uncertainty represented by the evidential grid. Real and simulated data were used to validate the planning algorithm with evidential grids.

Evidential-Based Approach for Trajectory Planning With Tentacles, for Autonomous Vehicles

“…8). In order to define a reward regarding the occupancy of the state, we propose to process cells information using three different rules [38], [31]. We consider that each cell is a source of information about the occupancy of the state.…”

“…Based on our previous results [30] [31], this paper presents a trajectory planning method combining the clothoids and MDP techniques with the use of evidential grids to integrate perception uncertainty. Then, it proves the performances of the approach by a global validation, under the simulator SCANeR™Studio.…”

A Two-Stage Real-Time Path Planning: Application to the Overtaking Manuever

“…Mouhagir et al proposed a method to deals with the uncertainty in the perception of the environment by combining Belief Functions to build an evidential occupancy grid and clothoid tentacles for trajectory planning on dynamic environments [26]. The evidential grid considers the safety distances between the automated vehicle and the obstacles, and they enhance the binary occupancy grid by including the road limit and the longitudinal expansion of the dynamic obstacles.…”

Lane level context and hidden space characterization for autonomous driving

“…However, classification and prediction of dynamic occupied cells is more complicated, grids are less suited to manage dynamic road users. The current trend is to use a prior map with several layers like a geometric layer, a topological layer and a semantic layer (see [7]) which can be used jointly with the detected objects as in [8] or with the free space characterized by an occupancy grid [9]. The environment representation is therefore a key point to understand the situation and to infer knowledge.…”