You can accomplish what you want by using rays cast against geometry to determine intersection points. Consider you have your player standing on the ground. You can cast a vertical ray downward and determine the intersection of that ray and the ground to find the spot to place the player object. This is basically what you are doing when you determine the elevation in a heightmap. But the same principle applies to non-heightmap geometry. For example, now say that your player takes a step. You can cast a ray downward at the location of where the step would take him, and calculate again the Y height (ray/polygon intersection calculation is fairly simple). This time, the ray intersects a ramp. The difference in Y between the player's current location and the new location is compared. If the difference is too large, then the player can not step there since it would require climbing too high. If the step is deemed to be allowed, then the player is moved to the new location.
In one of my older projects, I would perform movement and interaction in this way. Each time the player took a step, I would calculate the new X/Z location for the player, and calculate the new Y as the player's current Y plus his maximum Step-Up height. I would then use this new point as the origin of a downward ray, and calculate the Y intersection. If no intersection was found, this indicated that there was no geometry underneath the new point (which could occur if the new point was inside a wall and there was no floor below it), and the movement was disallowed. If an intersection was found, then I would construct a collision volume at the new location, and test the volume for any collisions. If there were collisions (which could occur if the new point lay inside a piece of geometry) then the move was disallowed.
As an aside, in later iterations I used the
Recast and Detour library to simplify the process by constructing a navigation mesh. The navigation mesh is constructed with parameters representing agent maximum Step-up and agent Height, and locations in the mesh where an agent could not possibly get to based on its height and step-up value are automatically excluded from the final mesh. The result is a mesh that is a sub-set of the original level area, representing all areas the agent can possibly get to, with functionality to determine the height location of a given input point, as well as built-in pathfinding functionality to determine a path between any two connectable points. Doing the agent-height culling of areas in the nav-mesh generation eliminates many of the agent->world collision tests, since if a location has too little headroom for the agent to stand there, that location is culled from the final nav-mesh.
