There have been interesting experiments on corvids which show their ability to "put themselves into another being's shoes", e.g.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2346514/ (first hit on Google):
> Food-caching corvids hide food, but such caches are susceptible to pilfering by other individuals. Consequently, the birds use several counter strategies to protect their caches from theft, e.g. hiding most of them out of sight. When observed by potential pilferers at the time of caching, experienced jays that have been thieves themselves, take further protective action. Once the potential pilferers have left, they move caches those birds have seen, re-hiding them in new places. Naive birds that had no thieving experience do not do so. By focusing on the counter strategies of the cacher when previously observed by a potential pilferer, these results raise the intriguing possibility that re-caching is based on a form of mental attribution, namely the simulation of another bird's viewpoint. Furthermore, the jays also keep track of the observer which was watching when they cached and take protective action accordingly, thus suggesting that they may also be aware of others' knowledge states.
The key to infering that the birds were "simulating" others' decisions was that their behaviour changed depending on whether the cache was spotted by another individual or not (it also depended on the birds' past experience of pilfering; but that only shows that the tactic is learned rather than innate).
Perhaps something similar could be done for empathy; for example, give the animals a mechanism to free themselves, which requires knowledge to operate (e.g. operating levers in a sequence). Replicate this experiment in the new setting, to ensure the animals, when shown the correct sequence, will reliably free themselves and others.
Now, go a little further; show the correct sequence to animals in pairs, where sometimes both can see the correct sequence, and other times only one can see (e.g. one can't see past a partition). Now those animals which know the sequence also know whether their partner knows or not. Once they've freed themselves, will they go on to help free the other? If they free those which could not see the sequence, but do not free those which could see the sequence, that would be evidence in favour of empathy, i.e. they're able to "put themselves into the other animal's shoes" in terms of knowledge/information ("he doesn't know how to escape, I'll help") and emotionally ("he could escape if he wanted to, so he must not want to").
This may require controlled behaviour, to engineer a situation where both animals could escape but only one does. Maybe a dummy/puppet/etc. could be used.