Agent Architecture for Adaptive Behaviors in Autonomous Driving

Lio, Mauro Da; Donà, Riccardo; Papini, Gastone Pietro Rosati; Gurney, Kevin

doi:10.1109/access.2020.3007018

Cited by 16 publications

(10 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The planned trajectory may also be updated when accumulated drifts exceed a given threshold. A study concerning the stability of the re-planning loop, for the agent architecture described in [61], is given in [56] (see also next section).…”

Section: Figure 12mentioning

confidence: 99%

“…The agent is described in [61] but, in order to understand the following section, some clarifications are given concerning trajectory planning and its interaction with trajectory execution. The agent elaborates action plans cyclically, in the form of two functions of time: the longitudinal control j(t) and the lateral control r(t).…”

Section: Evaluation In Closed Loopmentioning

confidence: 99%

“…This means that the agent also reacts to errors in the execution of the trajectory that might have happened in one cycle (initial position/orientation/curvature are updated). All deviations happening in one cycle are managed by the agent with the generation of a new motor plan that pursues the agent long-term goals from the deviated position, resulting in minimization of interventions (the "minimum intervention principle" [62], [61,Section II.D.3…”

Section: Evaluation In Closed Loopmentioning

confidence: 99%

“…At every cycle the agent select a trajectory among 1681 possible actions that correspond to lumping trajectories {j(t), r(t)} into bins that share the same initial control j(0) = j 0 , r(0) = r 0 . There are 41×41 bins, which have a neuralised tensor organisation [61,Section II.B]. The choice of one particular trajectory among the 41×41 candidates is based on their "salience" (a merit function).…”

Section: ])mentioning

confidence: 99%

“…There are two algorithms for action-selection: winner-takes-all (the trajectory with the momentarily instantaneous highest salience is chosen) and the multi-hypothesis sequential probability ratio test. The latter accumulates evidence of the most likely best choice on short observation times, which is more robust against motor and sensor noise [61,Section II.D]. In the following, the winner-takes-all algorithm is used because it does not hide/reduce the motor noise caused by inverse model errors.…”

Section: ])mentioning

confidence: 99%

See 4 more Smart Citations

A Mental Simulation Approach for Learning Neural-Network Predictive Control (in Self-Driving Cars)

et al. 2020

Self Cite

View full text Add to dashboard Cite

This paper presents a novel approach to learning predictive motor control via "mental simulations". The method, inspired by learning via mental imagery in natural Cognition, develops in two phases: first, the learning of predictive models based on data recorded in the interaction with the environment; then, at a deferred time, the synthesis of inverse models via offline episodic simulations. Parallelism with human-engineered control-theoretic workflow (mathematical modeling the direct dynamics followed by optimal control inversion) is established. Compared to the latter human-directed synthesis, the mental simulation approach increases autonomy: a robotic agent can learn predictive models and synthesize inverse ones with a large degree of independence. Human modeling is still needed but limited to providing efficient templates for the forward and inverse neural networks and a few other directives. One could consider these templates as the efficient brain network typologies that evolution produced to permit live beings quickly and efficiently learning. The structure of the neural networks-both forward and inverse ones-is made of interpretable "local models", which follows the cerebellar organization (and are also similar to local model approaches known in the literature). We demonstrate the learning of a first-round model (contrasted to Model Predictive Control) for lateral vehicle dynamics. Then, we demonstrate a second learning iteration, where the forward/inverse neural models are significantly improved.

show abstract

Section: Figure 12mentioning

confidence: 99%

Section: Evaluation In Closed Loopmentioning

confidence: 99%

Section: Evaluation In Closed Loopmentioning

confidence: 99%

Section: ])mentioning

confidence: 99%

Section: ])mentioning

confidence: 99%

See 3 more Smart Citations