California Partners for Advanced Transportation Technology

What some human drivers have done wrong has been blamed for much of the problem associated with the current highway systems. For example, driver inattention, fatigue and other human errors have often been cited as major sources of safety hazard on current highways and human capabilities as major limitations on current highway capacity. Such human deficiencies and the pervasive urban traffic congestion have motivated the concept of Automated Highway Systems (AHS). The fundamental objective of AHS is to achieve user and societal benefits through replacing human driving by automated machine driving. The first fundamental thesis of this paper is that safe replacement of human driving on highways by automation requires a rigorous examination into what most human drivers have been doing right on the current highway systems. Such an examination would provide much insight into the functional requirements for AHS, i.e., what machines must do to emulate or improve human driving. Furthermore, what human drivers tend to do poorly or well must be contrasted with what machines tend to do well or poorly. A complementary arrangement must be sought if the machines cannot safely or affordably replace driver intelligence for highway driving, either on a mature AHS or during intermediate stages toward it.

Driving tasks involved in any vehicle-roadway system depend on the driving environment. For example, city-street driving is different from driving on current highways and AHS driving may be drastically different from current highway driving. On an AHS, any human intelligence required by current highway driving must be replaced (emulated or improved) by machine intelligence or continues to be provided by the driver, if proven safe, unless the corresponding tasks are eliminated from AHS driving due to the environment change. Note that new driving environments may introduce new driving tasks. Since the driving environment during early deployment stages of AHS is likely identical to that of the current highways, most, if not all, of the current highway-driving tasks remain necessary. The second fundamental thesis of this paper is that current highway driving involves not only many routine chores, which machines tend to do well, but also much human intelligence, which machines tend to do poorly. Such intelligence is often required in subtle ways or is required only when unusual but normal events occur. We identify many such driving functions. If some of these required functions cannot be safely emulated by machines either in a mature end-state AHS or during intermediate stages toward the end state, then "driver-in-the-loop'' must not be ruled out at the current research and development stage.