Behavioral embeddings were calculated using 3D joint angles from the same 1 million frames used in Figure 4A. (A) Behavioral map combining all data during 4 s of optogenetic stimulation of MDN>CsChrimson (n = 6 flies, n = 29 trials), aDN>CsChrimson (n = 6 flies, n = 30 trials), and wild-type control animals (For MDN-Gal4/+, n = 4 flies, n = 20 trials. For aDN-Gal4/+ n = 4 flies, n = 23 trials). The same behavioral map is shown with only the data from (B) MDN>CsChrimson stimulation, (C) aDN>CsChrimson stimulation, or (D) control animal stimulation. Associated videos reveal that these distinct map regions are enriched for backward walking, antennal grooming, and forward walking, respectively.