Modeling Preemptive Behaviors for Uncommon Hazardous Situations From Demonstrations
ثبت نشده
چکیده
This paper presents a learning from demonstration approach to programming safe, 1 autonomous behaviors for uncommon driving scenarios. Simulation is used to 2 re-create a targeted driving situation, one containing a road-side hazard creating 3 significant occlusion in an urban neighborhood, and collect optimal driving behav4 iors from 24 users. Paper employs a key-frame based approach combined with 5 an algorithm to linearly combine models in order to extend the behavior to novel 6 variations of the target situation. This approach is theoretically agnostic to the kind 7 of LfD framework used for modeling data and our results suggest it generalizes 8 well to variations containing additional number of hazards occurring in sequence. 9 The linear combination algorithm is informed by analysis of driving data, which 10 also suggests that decision making algorithms need to consider a trade-off between 11 road-rules and immediate rewards to tackle some complex cases. 12
منابع مشابه
Alcohol expectancies and social self-efficacy as mediators of differential intervention outcomes for college hazardous drinkers with social anxiety.
The current pilot study examined the roles of two cognitive factors - positive alcohol expectancies of social anxiety reduction and drink refusal self-efficacy relevant to social situations - in mediating greater reduction in alcohol behaviors by the Brief Intervention for Socially Anxious Drinkers (BISAD; n=21) compared to an alcohol psychoeducation (n=20) in a sample of college hazardous drin...
متن کاملIdentification of Hazardous Situations using Kernel Density Estimation Method Based on Time to Collision, Case study: Left-turn on Unsignalized Intersection
The first step in improving traffic safety is identifying hazardous situations. Based on traffic accidents’ data, identifying hazardous situations in roads and the network is possible. However, in small areas such as intersections, especially in maneuvers resolution, identifying hazardous situations is impossible using accident’s data. In this paper, time-to-collision (TTC) as a traffic conflic...
متن کاملOne-Shot Imitation from Observing Humans via Domain-Adaptive Meta-Learning
Demonstrations provide a descriptive medium for specifying robotic tasks. Prior work has shown that robots can acquire a range of complex skills through demonstration, such as table tennis (Mülling et al., 2013), lane following (Pomerleau, 1989), pouring water (Pastor et al., 2009), drawer opening (Rana et al., 2017), and multi-stage manipulation tasks (Zhang et al., 2018). However, the most ef...
متن کاملThe Hazardous Situation Ontology Design Pattern
This extended abstract describes an ontology design pattern for modeling hazardous situations. We build upon state-of-art models for hazards and hazardous events, and on existing standards in the domain of occupational safety. We also present an example of the application of the pattern in the occupational safety and health domain.
متن کاملIntegrating Anticipatory Competence into a Bayesian Driver Model
We present the probabilistic model architecture for combining a layered model of human driver expertise with beliefs about the vehicle state to describe the effect of anticipations on driver actions. It implements the sensory-motor system of human drivers in a psychological motivated mixture-of-behaviors (MoB) architecture with autonomous, goal-based attention allocation and anticipation proces...
متن کامل