A novel and fun exercise robot (LOCOBOT) was developed to improve balance ability. This system can control a spherical robot on a floor by changing the center of pressure (COP) based on weight-shifting on a board. The present study evaluated leg muscle activity and joint motion during LOCOBOT exercise and compared the muscle activity with walking and sit-to-stand movement. This study included 10 healthy male adults (age: 23.0 ± 0.9 years) and examined basic LOCOBOT exercises (front–back, left–right, 8-turn, and bowling). Electromyography during each exercise recorded 13 right leg muscle activities. Muscle activity was represented as the percentage maximal voluntary isometric contraction (%MVIC). Additionally, the joint motion was simultaneously measured using an optical motion capture system. The mean %MVIC differed among LOCOBOT exercises, especially in ankle joint muscles. The ankle joint was primarily used for robot control. The mean %MVIC of the 8-turn exercise was equivalent to that of walking in the tibialis anterior, and the ankle plantar flexors were significantly higher than those in the sit-to-stand motion. Participants control the robot by ankle strategy. This robot exercise can efficiently train the ankle joint muscles, which would improve ankle joint stability.