Recent major advances in understanding the organizational principles underlying motor control have focused on a small number of animal species with stiff articulated skeletons. These model systems have the advantage of easily quantifiable mechanics, but the neural codes underlying different movements are difficult to characterize because they typically involve a large population of neurons controlling each muscle. As a result, studying how neural codes drive adaptive changes in behavior is extremely challenging. This problem is highly simplified in the tobacco hawkmoth Manduca sexta, which, in its larval stage (caterpillar), is predominantly soft-bodied. Since each M. sexta muscle is innervated by one, occasionally two, excitatory motor neurons, the electrical activity generated by each muscle can be mapped to individual motor neurons. In the present study, muscle activation patterns were converted into motor neuron frequency patterns by identifying single excitatory junction potentials within recorded electromyographic traces. This conversion was carried out with single motor neuron resolution thanks to the high signal selectivity of newly developed flexible microelectrode arrays, which were specifically designed to record from M. sexta muscles. It was discovered that the timing of motor neuron activity and gait kinematics depend on the orientation of the plane of motion during locomotion. We report that, during climbing, the motor neurons monitored in the present study shift their activity to correlate with movements in the animal's more anterior segments. This orientation-dependent shift in motor activity is in agreement with the expected shift in the propulsive forces required for climbing. Our results suggest that, contrary to what has been previously hypothesized, M.sexta uses central command timing for adaptive load compensation.