Human-Inspired Autonomous Vehicle Highway Navigation

Although many algorithms have been developed to plan a single shortest (or safest) path for a given environment, little progress has been made in effective shared human-machine control vehicles. The focus of this research is to design active navigation systems for autonomous or semi-autonomous vehicles based on the notion of "fields of safe travel" or "homotopy classes."

The motivation behind this approach is the assumption that human drivers tend to operate vehicles within fields of safe travel rather than rigidly follow a specific path. In other words, humans usually make decisions about high-level avoidance strategies or desired goals first, before determining a specific path of travel.

Based on this observation, we are developing a framework that relies on identification and analysis of candidate fields of safe travel. Such navigation system attempts to merge the homotopy selection process with a human driver's decision based on estimation of the driver’s intent. Once a desired field of safe travel has been identified, the system ensures vehicle safety while still allowing the human operator to have substantial control freedom.

Our approach to control system intervention is based on a measurement of threat to the vehicle. The autonomous control system ideally operates only during instances of significant threat: it should give a driver full control of the vehicle in "low threat" situations but apply appropriate levels of computer-controlled actuator effort during "high threat" situations.

This approach preserves the freedom of control of the human driver when he/she remains within a safe navigable corridor, and adjust the vehicle trajectory when its predicted future state falls out of a safe field, or when the lowest threat exceeds some threshold. In fully autonomous mode, this human-inspired motion planning approach ensures collision free navigation and driving comfort.

Three key issues related to the semi autonomous navigation approach are the following:

  1. Identification of fields of safe travel based on on-board vehicle sensory information (i.e. to identify nearby vehicles, pedestrians, road edges, etc.).
  2. Threat assessment, or characterization of the safety of each field of safe travel.
  3. The question of when and how to intervene based on threat, in a way to achieve both vehicle safety and respect of the operator's decision or preferences.

These three questions are related to one other; appropriate assessment of threat directly affects the performance of the system. In our current research, two different approaches are investigated: Approach A: threat assessment based on optimal trajectory analysis (i.e. workspace-based); Approach B: threat assessment based on margin of control freedom (i.e. input space-based).

In approach A we assesses the threat of the “best-case driving scenario.” The cost of the optimal trajectory is assumed to be the minimum required cost for ensuring vehicle safety, so the system intervenes only when the optimal maneuver has a high cost for achieving safety. The cost function could be designed to consider the severity of the optimal maneuver, through inclusion of longitudinal and lateral acceleration, sideslip angles, etc. Given a cost function, this work is investigating a divide-and-conquer strategy of utilizing a homotopy identification framework for efficient optimization.

In approach B, the threat to a vehicle is assumed to be proportional to the freedom of control afforded by a particular field. This implies that (for instance) drivers tend to prefer to navigate through regions that are wide and exhibit low curvature. To characterize threat in each candidate field, a set of feasible trajectories is evaluated (rather than a single trajectory as in the first approach). For this purpose, a lattice is constructed via sampling, either in input or state space. This sampling method enables identification of sets of feasible trajectories associated with each field; from which metrics related to control freedom can be derived.

Human Inspired Autonomous Driving System from Alexandre Constanin on Vimeo.

The first video shows an autonomous driving scenario where the host vehicle first slows down to avoid collision with lead vehicles. As soon as one of the lead vehicles accelerates, the host also accelerates to overtake the slow traffic.

Human-Inspired Autonomous Highway Vehicle Navigation from Alexandre Constanin on Vimeo.

The second video shows results from a human test subject driving in a simulator. Each homotopy candidate is represented by a sequence of discs. The volume of the disc is proportional to the control freedom within the corridor and the color depends on the cost of the optimal trajectory.