Methods for Stochastic Collection and Replenishment (SCAR) optimisation for persistent autonomy

Abstract

Consideration of resources such as fuel, battery charge, and storage space, is a crucial requirement for the successful persistent operation of autonomous systems. The Stochastic Collection and Replenishment (SCAR) scenario is motivated by mining and agricultural scenarios where a dedicated replenishment agent transports a resource between a centralised replenishment point to agents using the resource in the field. The agents in the field typically operate within fixed areas (for example, benches in mining applications, and fields or orchards in agricultural scenarios), and the motion of the replenishment agent may be restricted by a road network. Existing research has typically approached the problem of scheduling the actions of the dedicated replenishment agent from a short-term and deterministic angle. This paper introduces a method of incorporating uncertainty in the schedule optimisation through a novel prediction framework, and a branch and bound optimisation method which uses the prediction framework to minimise the downtime of the agents. The prediction framework makes use of several Gaussian approximations to quickly calculate the risk-weighted cost of a schedule. The anytime nature of the branch and bound method is exploited within an MPC-like framework to outperform existing optimisation methods while providing reasonable calculation times in large scenarios.

‘The ‘Vector-Model’ wavefunction: spatial description and wavepacket formation of quantum-mechanical angular momenta’

“…We note that the j-wavepackets are illustrated by plotting the population distribution ( ) q j Q , of the maximum projection state along the direction defined by the angles q j , [36][37][38][39][40][41]. We also note an effect of the finite physical range of m and j: when ¢ m and ¢ j in equations (3) and (7) take non-physical values, we clamp the Gaussians to zero and the distributions become rectified [42]. This affects the statistical properties of the distributions and shifts the polar orientation angle of the wavepackets, which depends on the mean value of m. We use corrective formulas to account for this effect (see [42] and appendix B), with good results, particularly for  Dm 3, as shown in figure 2(f).…”

Multi-trainer Interactive Reinforcement Learning System

“…For example, if a trainer's real trustworthiness is 80%, he has 80% probability to give a correct answer. The trainers' real trustworthiness is generated by extended rectified Gaussian distribution with different mean and standard deviation [18]. To simulate the different trust distributions, we designed 6 groups of experiments with different standard deviations of trainers' trustworthiness (0, 0.1, 0.2, 0.3, 0.4, 0.5).…”

The Dynamic Bowser Routing Problem

“…Moving our attention to robotics, recently [49,50,51] investigated stochastic collection and replenishment of agents motivated by use cases in mining and agricultural settings in which a replenishment agent transports a resource between a centralised replenishment point to agents using the resource in the field. They employ Gaussian approximations to quickly calculate the risk-weighted cost of a schedule; a branch and bound search then exploits these predictions to minimise the downtime of the agents.…”

“…However, our modeling and solution framework has the advantage of relying solely on MILP modeling and not on ad-hoc algorithms to predict future asset resource levels. Moreover, the discussion in [51] assumes that uncertain parameters and variables are normally distributed, while our approach based on piecewise linearisation of loss functions does not require this assumption and can accommodate any distribution.…”