` ALAN: Assembly Line Adaptive "Netbot" (ALAN)

ALAN: Assembly Line Adaptive Netbot (ALAN):
A Practical Robot for Industry ?

Concept Sketch by Ken Goldberg (January 2009)

This note describes a concept tentatively called ALAN: Assembly-Line Adaptive Netbot, a proposed practical robot system for applications in manufacturing, material handling, and food production that is emerging from discussions between leaders from industry and academic research.

1) Industry expert report that the major obstacle to use of robots is the collateral costs and time required to design and create "robot compatible" environments with feeders, fixtures, lighting, sensors, etc. These costs can exceed the cost of robot arms by a factor of 2x-10x.

2) Consider a hypothetical alternative: the ALAN, Assembly-Line Adaptive Netbot: a practical robot system with the approx. form-factor and basic manipulation ability of a human worker (two arms, some sensing). ALAN need not be "humanoid" in appearance, but is designed to be quickly installed in unmodified factory environments and rapidly "trained" using vision and local tele-operation to perform a given repetitive manipulation task. ALANs may have access to CAD models of parts. Once trained at a particular task, ALANs can store this information online for future access. ALANs would use machine learning to improve performance at any task over time and will be connected to the network to share data and call for human assistance when necessary.

3) ALANs would have human-safe features and so could be co-located with people without requiring changes to existing environments and would allow industry to adopt robots in stages over time. Industry experts say that such a robot at a cost of $100K would be appealing and have a payback period of less than two years.

4) ALANs could build on emerging advances in robot learning, statistical pattern recognition, laser range sensors, algorithmic automation methods.

To achieve this goal, research is needed in robot learning, robust manipulation, machine vision, mechanical modeling, motion planning, and related subfields.

A relatively low-cost hardware platform would allow research groups to evaluate experimentats and share results.

Related Links