Traumatic SCIs have long-term health, economic and social consequences, stressing the urgency to develop interventions to improve recovery after such injuries. Today, the only proven effective interventions to enhance recovery after SCI are activity-based rehabilitation therapies, such as locomotor training. However, locomotor training shows no or very limited efficacy to improve function after a severe SCI that induces paralysis of the limbs. To mimic the outcome of severe but incomplete SCI in rodents, we developed a model of double opposite-side lateral hemisections termed staggered hemisection in adult rats. This model induced permanent paralysis below the level of injury but leaves an intervening gap of intact neural tissue that provides a substrate for recovery. We showed that this SCI leads to degradation of motor functions, which correlates with the formation of aberrant neuronal connections below the lesion. Robotic devices with a rehabilitative purpose should act as propulsive or postural neuroprosthesis allowing training under natural conditions. Our versatile robotic interface provides multidirectional bodyweight support during overground locomotion in rats. We next evaluated the effects of robot-assisted gait training enabled by electrochemical stimulation of spinal circuits to restore locomotion after staggered hemisection SCI. We found that after two months of daily training, paralyzed rats recovered the ability to initiate, sustain and adjust bipedal locomotion while supported in the robot under electrochemical stimulation. This recovery correlated with ubiquitous reorganization of corticospinal, brainstem, and intraspinal fibers. We next evaluated whether this treatment was capable of restoring supraspinal control of locomotion after a clinically relevant SCI. Rats received a severe contusion of the spinal cord that spared less than 10% of intact tissue. Robot-assisted rehabilitation restored weight-bearing locomotion in all the trained rats when stimulated electrochemicallay and in a subset of rats in the absence of any enabling factors which paralelled with the reorganization of axonal projections of reticulospinal fibers below the contusion. Virus-mediated silencing of reticulospinal neurons projecting to lumbar segments demonstrated that these inputs were necessary to initiate and sustain walking after training. When delaying the onset of training by two months, in the chronic stage, all the rats regained voluntary locomotor movements but the extent of the recovery was reduced compared to rats trained early after SCI. The results provide a strong rationale to evaluate the impact of neuroprosthetic training to improve motor functions in human patients with incomplete SCI. Translation of treatment paradigms developed in rodent models into effective clinical applications remains a major challenge in biomedical research. Here, we studied recovery of motor functions in more than 400 quadriplegic patients who presented various degree of spinal cord damage laterality. We found that recovery increases with the asymmetry of early motor deficits. We conclude that emergence of spinal cord decussating corticospinal fibers and bilateral motor cortex projections during mammalian evolution supports greater recovery after lateralized SCI primates compared to rodents. Novel experimental models and dedicated therapeutic strategies are necessary to take advantage of this powerful neuronal substrate for recovery after SCI.