At DC, 25mA through 1.5K requires 37Vdc.
At AC if you want a full scale current reversal, ie peak positive(+25mA) current to peak negative (-25mA) current, in 2ms, that's 25 A/Second
So as V = L di/dt with your 2H inductance,you'll need 50V to do that (50 = 2 * 25)
Add the two together and we get 87V which makes me think that with an OTS +-50V amplifier, you'll just about be able to meet your control bandwidth requirement. So the question becomes, how important is a full scale drive swing in 2ms? Is that a number that's estimated, picked from thin air /guessed, or established via validated simulation and calculation etc??
Assuming that the actual "project" is to make the robot do what you want to achieve, getting distracted down a rabbit hole of DIY amplifiers seems, well, like a massive distraction to me. Sure, it'll be fun and interesting but this is why many projects never get finished ime...... (you start a project to build a robot, you end up with a custom power amplifier for an application no one else needs......

) Been there, bought that Tshirt.......
