Intelligent Robots in 2026: Are We There Yet? with Nikita Rudin

The discussion outlines the current state and future potential of legged robotics, emphasizing significant advancements in teaching robots to walk and navigate challenging environments through simulation and reinforcement learning. However, challenges remain in achieving high reliability and adaptability in real-world settings, primarily due to the sim-to-real gap and complex perception-based navigation. Researchers are exploring both end-to-end deep learning models and modular control approaches, with the latter showing greater practicality in the short term. The conversation highlights the difficulties in translating simulation-based learning to real-world environments, stressing the need for robust reward functions and semantic understanding. It also addresses the challenges of deploying robots in industrial vs. home settings, the role of imitation learning and reinforcement learning, and the use of large vision-language models in complex task orchestration. Despite rapid progress, translating laboratory success to reliable real-world applications remains a key challenge in robotics development.