Are you asking about the control system or just the gyros? Why are you interested in torque? If you want to apply counter torque with a motor you'll need to transform your motor's torque into an angular rate model to counter the induced rotation. Since you spent months on this it would be helpful to know where you are stuck and what methods you're using and what isn't working. I'll try to briefly cover both in general.
Micro Electrical Mechanical System gyros (MEMS) are not as accurate as high end ring laser gyros. MEMS have measurement errors which when integrated over time will lead to an accumulating error in the calculated orientation. Therefore, MEMS gyroscopes alone cannot provide an absolute measurement of orientation. And correcting for errors in measurement are important.
$\vec{\omega}_{measured}=\vec\omega_b\times\vec\omega_{sfcc}+\vec\omega_{bias}+\vec{Gs}\times\vec\omega_{gs}+\vec{noise}$
Where $\vec\omega_b$ is the body angular rates, $\vec\omega_{sfcc}$ is a 3x3 matrix of scaling factors on the diagonal and misalignment terms in the nondiagonal, $\vec\omega_{bias}$ are the bias values. $\vec{Gs}$ and $\vec\omega_{gs}$ are the gravity-sensitive biases.
If you just want to start by focusing on the gyroscopes, I suggest looking at this model which deals with the complexities of MEMS errors.
For stabilization you have to look at your inertial reference frame. An accelerometer and a magnetometer will measure the earth's gravitational and magnetic fields respectively and so provide an absolute reference of orientation. These devices are subject to high levels of noise; for example, accelerations due to motion will corrupt measured direction of gravity. So to make a Inertial Measurement Unit (IMU) you'll need an algorithm that integrates gyros, accelerometers and a magnetometer for best long term stabilization. Single estimates are made by filtering positional estimates through a kalman filter which reduces the average multiple errors in the various sensors. The kalman estimation also can help reduce gyroscopic bias drift.
All of these devices together including the filtering look like the following block diagram. Sebastian Madgwick developed an IMU and AHRS sensor fusion algorithm. Where $\hat{q}$ represents a quaternion describing the orientation of the earth frame relative to the sensor frame.
There are other parts that need some correction. Magnetic distortion needs to be corrected and more importantly gyroscope drift. The gyroscope zero bias will drift over time, with temperature and with motion and there are several different algorithmic choices in the literature to reduce this drift bias.
I suggest you read Madgwicks paper to really understand the dynamic aspects of IMU control. In addition here is his C source code or a Matlab version.
There is also a FreeIMU group dedicated to algorithmic improvements and open source code.
For the Arduino cpu based system there are some simpler tutorials like Bill Premmerlani's and Mahoney's papers.
EDIT
You need some sort of feed back control system to your motor. You don't need to worry about gravity or other forces as long as your motor has enough torque to stabilize movements of the platform. Most systems can be stabilized using a Proportional Integral Derivative (PID) controller:
In this drawing the "Process" is your motors responding to the correction voltage and the output are your gyros which should be showing no movement when stable.
Here is a good tutorial on how PID's work and how to optimize the parameters. And a wiki page on tuning the loop. Part of this involves measuring and characterizing your motor's response to voltage inputs so that you can set the PID gains properly for low-jitter and fast corrections.
However without an accelerometer the bias drift of your gyros will over time cause your platform to fall down. And at that you'll need to expand your system to the larger model I illustrated earlier which is very robust, but might be beyond your project's goals/time.
