This paper  presents a qualitative approach  to free space detection  for a
vision-based  autonomous vehicle.  Image  sequences  are processed  through
a  multiresolution computational  paradigm,  driven  by a-priori  knowledge
through  a top-down  control,  without  any kind  of  motion analysis.  The
algorithm,  running  on  a  special-purpose  low  cost  massively  parallel
architecture has been tested on board an autonomous robot.