2018 May 17 – Birbaumer and colleagues published in 1999 a pivotal paper on brain-computer interfaces (BCIs), showing subjects suffering from advanced amyotrophic lateral sclerosis use slow cortical potentials of the electroencephalogram to drive a cursor on a video screen, thus operating a spelling software. Over the years, BCI studies have largely ignored the contribution of user-training strategies while focusing on machine learning. However, the prizewinning approach in the 2016 Cybathlon BCI race benefited from the refinement of a mutual, incremental interaction between user training and machine learning. From the users’ perspective, this co-adaptive system allowed typical motor outputs to occur, in that these became automatic with practice (i.e. based on implicit learning, not on explicit information such as motor imagery). The related case study analyzed the two participants involved in this specific experimental set-up, who were both severely impaired as a consequence of spinal cord lesions at level C5–C6. They were trained to control their avatar by modulating sensorimotor rhythms in the μ (8–12 Hz) and β (18–30 Hz) ranges. Recordings where acquired by means of a 16 electrodes electroencephalographic cap targeting sensorimotor cortices.