Progress in Embodied AI has made it possible for end-to-end-trained agents to navigate in photo-realistic environments with high-level reasoning and zero-shot or language-conditioned behavior, but evaluations and benchmarks are still dominated by simulation. In this work, we focus on the fine-grained behavior of fast-moving real robots and present a large-scale experimental study involving 262 navigation episodes in a real environment with a physical robot, where we analyze the type of reasoning emerging from end-to-end training. In particular, we study the presence of realistic dynamics which the agent learned for open-loop forecasting, and their interplay with sensing. We analyze the way the agent uses latent memory to hold elements of the scene structure and information gathered during exploration. We probe the planning capabilities of the agent, and find in its memory evidence for somewhat precise plans over a limited horizon. Furthermore, we show in a post-hoc analysis that the value function learned by the agent relates to long-term planning. Put together, our experiments paint a new picture on how using tools from computer vision and sequential decision making have led to new capabilities in robotics and control.
Reasoning in visual navigation of end-to-end trained agents: a dynamical systems approach
| Steeven Janny, Hervé Poirier, Leonid Antsfeld, Guillaume Bono, Gianluca Monaci, Boris Chidlovskii, Francesco Giuliari, Alessio Del Bue, Christian Wolf |
| The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville, Tennessee, USA, 11-15 June, 2025 |
Dynamical Agent
The end-to-end agent interacts with a dynamical model for realistic motion.
We reproduce the work from Bono et al. and substantially improved their performance through longer training time, better angle representation and data augmentation.
Simulated dynamics greatly improves navigation in real world.
| Method & action space | +dyn (train) | Sim (+dyn.) HM3D/2.5k | Sim (+dyn.) office | Real office | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| SR | SPL | SCT | SR | SPL | SCT | SR | SPL | SCT | ||
| D4 | No | 29.1% | 18.1% | 2.0% | 10.0% | 7.3% | 0.9% | 0.0% | 0.0% | 0.0% |
| D28-instant | No | 27.6% | 11.6% | 5.0% | 25.0% | 11.1% | 5.8% | 10.0% | 5.3% | 1.7% |
| D28-dynamics | Yes | 97.6% | 82.3% | 52.5% | 100.0% | 82.1% | 64.5% | 92.5% | 61.1% | 22.4% |
Prediction vs. Correction
To navigate effectively, an agent can either rely solely on sensory inputs or use a learned dynamics model, integrating each step with feedback from prior actions. It’s believed that trained navigation agents combine both approaches through a prediction-correction process similar to a Kalman filter, where predictions from a dynamics model are adjusted with sensory feedback.
Dynamics agent uses a prediction-correction scheme to estimate its current pose.
Occupancy map
We probed the existence of local occupancy map in a square area of 3×3 meters around the agent.
The internal state contains the local occupancy map of the environment.
Action rating in real navigation
We used an expert to rate the actions of the agent in real navigation episodes based on Fast Marching Method. By observing how the scores changes from one step to another, we can identify difficult part of the path. Moreover, we can aggregate this information spatially and discover challenging areas within the environment.
Real robot navigates well, but struggles on narrow paths and dead ends.
Shapley values
We use Shapley values to understand the importance of each inputs in the agent’s decision-making process. Shapley values are a method from cooperative game theory that assigns a value to each player in a game based on their contribution to the game’s outcome.
Policy relies mostly on pose sensing and structured vision to navigate.
To make robots autonomous in real-world everyday spaces, they should be able to learn from their interactions within these spaces, how to best execute tasks specified by non-expert users in a safe and reliable way. To do so requires sequential decision-making skills that combine machine learning, adaptive planning and control in uncertain environments as well as solving hard combinatorial optimisation problems. Our research combines expertise in reinforcement learning, computer vision, robotic control, sim2real transfer, large multimodal foundation models and neural combinatorial optimisation to build AI-based architectures and algorithms to improve robot autonomy and robustness when completing everyday complex tasks in constantly changing environments.
The research we conduct on expressive visual representations is applicable to visual search, object detection, image classification and the automatic extraction of 3D human poses and shapes that can be used for human behavior understanding and prediction, human-robot interaction or even avatar animation. We also extract 3D information from images that can be used for intelligent robot navigation, augmented reality and the 3D reconstruction of objects, buildings or even entire cities.
Our work covers the spectrum from unsupervised to supervised approaches, and from very deep architectures to very compact ones. We’re excited about the promise of big data to bring big performance gains to our algorithms but also passionate about the challenge of working in data-scarce and low-power scenarios.
Furthermore, we believe that a modern computer vision system needs to be able to continuously adapt itself to its environment and to improve itself via lifelong learning. Our driving goal is to use our research to deliver embodied intelligence to our users in robotics, autonomous driving, via phone cameras and any other visual means to reach people wherever they may be.
This web site uses cookies for the site search, to display videos and for aggregate site analytics.
Learn more about these cookies in our privacy notice.
Cookie settings
You may choose which kind of cookies you allow when visiting this website. Click on "Save cookie settings" to apply your choice.
This content is currently blocked. To view the content please either 'Accept social media cookies' or 'Accept all cookies'.
For more information on cookies see our privacy notice.