While most (semi)autonomous mobile manipulators employ expensive articulated arms with grippers (6 or more DOF), the Healthcare Robotics Lab at Georgia Tech, the same folks who made EL-E, are also examining low-complexity end effectors modelled off of dustpans and kitchen turners for non-prehensile grasping of isolated objects from the floor. When mounted on an iRobot Create (Roomba), the system's performance was impressive; it successfully grasped ~95% of the 34 test objects across numerous orientations / configurations and four different surfaces -- an impressive feat of robustness given that the end effector is a single under-actuated "sweeper" (1 DOF) working in tandem with a planar wedge, the whole system operates via open loop control, and the objects were quite varied (from small individual pills to large containers, and from deformable textiles to rigid bottles). This system is slated to appear at ICRA 2009 in Kobe, Japan in the next few days and is documented in a paper entitled "1000 Trials: An Empirically Validated End Effector that Robustly Grasps Objects from the Floor" (of which I am a coauthor). Read further for videos and additional discussion.
A video of the robot in action is most illustrative.
The operating principal of this end effector is similar to a dustpan (plus broom) or a kitchen turner, illustrated below. In both cases, the object (quarter or waffle) is originally located on a planar surface. As a planar wedge (dustpan or kitchen turner) is advanced forward, an opposing force is applied either by a member (broom) or friction such that the object is "swept" onto the planar wedge. Ideally, this operation is quasi-static such that the object's initial configuration (orientation and state) is preserved after being transferred onto the planar wedge.
These design elements are readily visible in the design of the "Dustpan" end effector, which actually incorporates a kitchen turner for part of the wedge.
The video showed the robot performing just a few of the total grasp attempts. For a more comprehensive look at the robot's performance, I would recommend reading the ICRA 2009 paper, 1000 Trials: An Empirically Validated End Effector that Robustly Grasps Objects from the Floor.
Finally, the paper alludes to a revised "Dustpan" end effector system that has (among other things) a scissor lift to presumably allow full object retrieval: grasping followed delivery (a common theme in the Healthcare Robotics Lab). The new system is still in the very early stages, but statically posed images of the new robot are certainly enticing.