I am an employee of DSTA, but I do not speak in my capacity as one.
The purpose of such contests is typically not to field an operational capability. It is very unlikely that the winning robot or a variant will actually be deployed. The main purpose is to encourage industry and academia to perform research in certain fields, such as machine vision, control systems, AI, etc. This is a long term investment. The secondary purpose is to gauge the state of the art in these fields while advancing it. This is the short term gain.
The contest is modelled after the DARPA Grand Challenge, which concentrates on outdoor navigation. Similarly, you will not see autonomous combat vehicles anytime soon. However, DARPA has certainly focussed interest and effort toward all the fundamental research questions needed to achieve such a feat. DARPA also now has a good idea of what is possible when planning acquisitions and upgrades, and is able to better assess the technical risk of new developments. If the US Army asked for an autonomous UGV tomorrow, DARPA would be able to give a good estimate of how much it would cost, how long it would take, and what is realistically achievable (then the politicians will come in and screw things up).
Such contests are an admission that the state of the art is no longer in the military or intelligence communities, but in the acadamic and industrial spheres. AES was developed outside the NSA, for example. More and more equipment is COTS or MOTS (commercial / militarized off-the-shelf). The days when you could get a national laboratory (Singapore has one too) to singlehandedly advance the state of the art are long over. Nowadays inhouse research tends to be focussed on either security-sensitive fields, or areas no one else simply wants to touch. This trend will only accelerate in the future.