Acceleration-based policy definition
- A Policy now computes an acceleration instead of a new velocity. This is more in line with the I3D 2020 paper.
- The role of the agent mass has changed, and the default mass is now 1.
- The agent's acceleration and velocity are now clamped at a more logical point in the simulation.
- Added CostFunction::GetGradientFromCurrentVelocity(), which can improve efficiency.