Planning the sequence of components (or parts) to be assembled during manuf
acturing is an important application problem for virtual environments for t
hree main reasons. First, it is a difficult combinatorial optimization but
a highly visual problem. Second, a majority of assembly operations in facto
ries (with the exception of simple pick-and-place operations) are still per
formed manually, because they are difficult to automate. Hence, it is an im
portant problem involving human-machine interface. Third, there are a numbe
r of assembly operations which require dextrous operator training. Hence, i
t is also an important training problem. Recent research suggests a promisi
ng approach for assembly determination based on using heuristic rules to ge
nerate soft constraints in addition to the regular hard quantitative constr
aints due to part geometry and topology. We believe that the emergence of v
irtual environments can enable us to systematically use these soft constrai
nts, which previously has not been possible. In this paper, we report resul
ts of experiments involving fifteen voluntary participants using a nonvirtu
al reality (VR) environment involving blueprints, a nonimmersive desktop VR
environment, and an immersive projection-based VR environment to first tea
ch participants skills in handling soft and hard constraints for assembly p
lanning through examples, and then to measure the effectiveness of their le
arnt skills in solving a different example problem. We have classified soft
constraints as infeasibility constraints, reorientation constraints, diffi
culty constraints, instability constraints, and dissimilarity constraints.
A significant observation is that the participants could, on average, perfo
rm the assembly operations in approximately half the time in the immersive
and nonimmersive VR environments than in the traditional environment using
blueprints.