Intelligent Robots in 2026: Are We There Yet? with Nikita Rudin - #760

2026-01-08 · Source: The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, extended

Summary

Nikita Rudin, co-founder and CEO of Flexion Robotics, discusses the current state of robotics, highlighting the significant gap between existing capabilities and the requirements for fully autonomous real-world deployment. He explains that while reinforcement learning and simulation have advanced robot locomotion, particularly for "blind" quadruped robots, adding visual inputs introduces noise and complicates sim-to-real transfer. The discussion covers the pragmatic benefits of modular approaches over end-to-end models for planning and locomotion, and Flexion's hierarchical strategy using pre-trained Vision-Language Models (VLMs) for high-level task orchestration with Vision-Language-Action (VLA) models and low-level whole-body trackers. Rudin also introduces "real-to-sim" for refining simulation parameters and shares insights into the challenges of humanoid robot demos, reward tuning, and the future of robotics in industrial and home settings.

Key takeaway

For AI Scientists and Research Scientists developing autonomous robots, prioritize closing the sim-to-real gap, especially when incorporating visual perception. Your efforts should focus on modular architectures that leverage pre-trained VLMs for high-level reasoning, while refining low-level control through robust simulation and real-to-sim data. This approach will accelerate the transition from controlled demos to reliable, value-generating deployments in industrial settings, before tackling the complexities of home environments.

Key insights

Bridging the sim-to-real gap and integrating perception are key challenges for deploying autonomous robots.

Principles

• Locomotion is not "solved" until robots can reliably navigate any human-accessible terrain.
• Modular approaches can pragmatically overcome end-to-end training challenges in complex robotics.
• High-fidelity simulation requires deep understanding of both simulated and real-world robot physics.

Method

Flexion Robotics employs a hierarchical approach: VLMs orchestrate high-level tasks, VLAs plan kinematic motions, and whole-body trackers control motors. This combines RL, imitation learning, and teleoperation data, refined by "real-to-sim" processes.

In practice

• Use pre-trained VLMs for high-level task orchestration in complex robotic systems.
• Employ "real-to-sim" to refine simulation parameters using real-world robot data.
• Start with simpler, repetitive industrial tasks for early humanoid robot deployment.

Topics

• Reinforcement Learning
• Sim-to-Real Transfer
• Humanoid Robotics
• Vision-Language Models
• Robot Locomotion

Best for: AI Scientist, Research Scientist, Robotics Engineer, Machine Learning Engineer, AI Researcher

Related on AIssential

• Intelligent Robots in 2026: Are We There Yet? [Nikita Rudin] - 760 — Bridging the sim-to-real gap and integrating semantics are critical for advanced robot autonomy and value generation.
• Training a Unitree G1 to Walk w/ Reinforcement Learning — Achieving sim-to-real transfer for robot locomotion requires meticulous sim-to-sim fidelity and addressing hardware-specific observation gaps.
• This simulation startup wants to be the Cursor for physical AI — High-fidelity simulation is crucial for scaling physical AI and bridging the gap between virtual training and real-world robot deployment.
• From Seeing to Simulating: Generative High-Fidelity Simulation with Digital Cousins for Generalizable Robot Learning and Evaluation — Digital Cousins generates diverse, high-fidelity simulations from real-world panoramas for robust robot learning.
• What is Physical AI? How Robots Learn & Adapt in Real Life — Physical AI integrates perception, reasoning, and action in the physical world, driven by advanced models and efficient compute.
• NVIDIA Research Advances Robotics From Simulation to the Real World — Simulation-to-real transfer is foundational for achieving generalizable, reliable embodied autonomy in robotics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence).