In this paper

we propose a path selection scheme based on the idea of autonomous learning for concurrent multipath transfer (CMT) in path selection.Firstly

this autonomous learning scheme integrates the requirements of a specific service

such as throughput

delay and loss rate

to define the optimization object function.Secondly

this scheme sets several parameters of path selection as the learning parameters.During the transmission

through analyzing the relationship between different values of learning parameters and the actual measured values of object function

the system performs its autonomous learning

and gradually gets the best value combination of learning parameters.Experimentations show the feasibility

convergence and stability of the scheme

and also demonstrate the scheme could adjust the learning parameters by itself to adapt the changing network condition and gain the best value of object function.