To　make　a　driverless　car　with　better　environment　awareness,　multi-layer　laser　radar　was　applied　to　detect　roads　and　obstacles.　Firstly　the　road　edge　data　set　was　extracted　from　numerous　laser　radar　data　based　on　characteristics　of　the　road　edge　data,　and　the　cluster　analysis　of　the　data　sets　was　done　with　the　improved　COBWEB　algorithm　based　on　Euclidean　distance.　In　order　to　divide　the　road　into　drivable　area　and　undrivable　area,　the　left　and　right　road　edges　were　respectively　fitted　into　a　straight　line　with　the　least　squares　method.　Secondly　DSmT　was　applied　to　establish　a　grid　map　for　the　environment,　and　dynamic　obstacles　were　detected　by　the　conflict　coefficient　within　drivable　area.　Finally,　the　cluster　analysis　and　information　extraction　of　dynamic　obstacles　was　completed　by　the　expansion　algorithm,　erosion　algorithm　and　improved　eight　neighborhood　labeling　algorithm.　The　results　show　that　the　algorithm　can　significantly　reduce　redundant　operations　and　improve　efficiency.　©　2015　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.