Traversability analysis is a key step for ground robots to effectively navigating forest environments. Exteroceptive data provides rich information about the surroundings, but is often insufficient to achieve the desired navigational robustness. To address this limitation, proprioceptive data has been used lately in combination with exteroceptive sensor modalities. This work presents an ablation study performed with a self-supervised learning-based traversability technique that combines exteroceptive and proprioceptive data to generate pixel-wise traversability predictions on input RGB images. Performance with and without proprioception is assessed, as well as with different proprioceptive supervision signals, both qualitatively and quantitatively. Results show that introducing proprioceptive feedback does indeed improve the quality of monocular-based traversability analysis. Furthermore, the proprioceptive signal chosen can also play a relevant role in the quality of the final predictions.