New Capabilities for the Four-Legged KAIST HOUND Robot
Without any operator input, the KAIST HOUND quadruped robot has learned to shift its running style—including trotting and leaping—to overcome obstacles. This breakthrough is powered by artificial intelligence, specifically a reinforcement learning system called APT-RL. Weighing 45 kilograms, the robot uses cameras and lidar to scan its surroundings. Such advances highlight how AI-driven robots are becoming more adaptable in real-world settings.
Testing and Achievements
During outdoor trials, KAIST HOUND successfully navigated a 1.1-kilometer route through a university campus and a 0.3-kilometer forest path filled with roots, logs, and slippery leaves. The researchers built a training platform called “Action-Pre-trained Transformer Reinforcement Learning” (APT-RL). Training began with a simple 2D computer model of the robot, resulting in a dataset of 180,000 short sequences of trots and jumps—equivalent to about 15.5 hours of movement. Generating this dataset took roughly eight minutes.
In one indoor test, KAIST HOUND cleared a 60-centimeter-high obstacle and briefly reached speeds of up to 15 km/h. The robot also demonstrated the ability to jump down three-step staircases. However, its current design only supports two gait patterns and primarily handles forward motion.
Meanwhile, Chinese company Robbyant unveiled two new AI models: LingBot-Depth 2.0 and LingBot-Vision. LingBot-Depth 2.0 was trained on 150 million samples and achieved top scores in 12 out of 16 tests for scene depth recovery. The KAIST HOUND quadruped robot’s ability to navigate complex real-world environments—like stairs—marks significant progress in robotics and artificial intelligence.
Advancements in robotics, especially for quadrupedal machines, open new possibilities in fields ranging from rescue operations to exploration in challenging conditions. Using AI to train these robots allows them to adapt to new challenges and interact effectively with their surroundings, boosting their autonomy. This, in turn, could drive the development of technologies that make life easier for people and improve task efficiency in complex situations.
In addition to advancements in quadrupedal robots, the field of robotics is also witnessing innovations in humanoid designs. For instance, the introduction of the humanoid robot R-Noid by Robot.com marks a significant step forward in industrial automation, showcasing the diverse applications of AI-driven machines in various environments.