Context-Conditional Navigation with a Learning-Based Terrain- and Robot-Aware Dynamics Model

TL;DR

TRADYN addresses navigation under both terrain- and robot-parameter variability by learning a context-conditioned forward dynamics model that can adapt online. It extends a Neural Processes-based framework to include a terrain-dependent term ${\alpha}_{terrain}({\mathbf{x}})$ and a latent robot context ${\boldsymbol{\beta}}$, incorporating terrain features ${\boldsymbol{\tau}}$ via a GRU-based predictor. The model is trained with an ELBO objective and calibrated online through context transitions, enabling model-predictive control with terrain-aware planning using a terrain lookup. In 2D simulations, TRADYN achieves lower long-horizon prediction errors and reduced throttle energy compared to ablations, demonstrating practical benefits for robust, adaptive navigation in varying environments.

Abstract

In autonomous navigation settings, several quantities can be subject to variations. Terrain properties such as friction coefficients may vary over time depending on the location of the robot. Also, the dynamics of the robot may change due to, e.g., different payloads, changing the system's mass, or wear and tear, changing actuator gains or joint friction. An autonomous agent should thus be able to adapt to such variations. In this paper, we develop a novel probabilistic, terrain- and robot-aware forward dynamics model, termed TRADYN, which is able to adapt to the above-mentioned variations. It builds on recent advances in meta-learning forward dynamics models based on Neural Processes. We evaluate our method in a simulated 2D navigation setting with a unicycle-like robot and different terrain layouts with spatially varying friction coefficients. In our experiments, the proposed model exhibits lower prediction error for the task of long-horizon trajectory prediction, compared to non-adaptive ablation models. We also evaluate our model on the downstream task of navigation planning, which demonstrates improved performance in planning control-efficient paths by taking robot and terrain properties into account.

Figures (8)

• Figure 1: Terrain- and robot-aware control-efficient navigation. We propose a method for control-cost optimal navigation with learned dynamics models. Our method can adapt to varying, unobserved properties of the robot, such as the mass, and spatially varying properties of the terrain, such as the friction coefficient. In the above example of navigating from a single starting point (white cross) to two different goals (black cross), as a result, our method circumvents areas of high friction coefficient and favors areas of low-friction coefficient. As the dissipated energy also depends on the mass of the robot, a heavy robot ($m= 4kg$, blue, orange) is allowed to take longer detours to the goal than a light robot ($m= 1kg$, green, red).
• Figure 2: Architecture of our proposed terrain- and robot-aware forward dynamics model (TRADYN). The initial state of the robot $\bm{x}_0$ is embedded as hidden state of a gated recurrent unit (GRU) cell. The GRU makes a single-step forward prediction in the latent space using embeddings of the context variable $\bm{\beta}$, action $\bm{u}$ and terrain observation $\bm{\tau}$ as additional inputs. Latent states are mapped to Gaussian distributions on the robot's observation space for decoding. While during training the actual terrain observation ${\tau}(\bm{x}_n)$ is used, during prediction, the map ${\tau}$ is queried at predicted robot locations ${\tau}({\color{red}\bm{\hat{x}}_n})$. See \ref{['sec:tradyn:method']} for details.
• Figure 3: Exemplary rollouts (length 50) on two different terrain layouts (rows) and for two exemplary robot configurations (low-inertia, high-inertia) (columns). Rollouts start from the center; actions are sampled time-correlated. The low-inertia robot has minimal mass $m=1$ and maximal control gains $k_\mathrm{throttle}=1000$, $k_\mathrm{steer}=\pi/4$. The high-inertia robot has maximal mass $m=4$ and minimal control gains $k_\mathrm{throttle}=500$, $k_\mathrm{steer}=\pi/8$. Equally colored trajectories (, , ) correspond to identical sequences of applied actions. See \ref{['sec:tradyn:simenv']} for details.
• Figure 4: Relationship of RGB terrain features $\bm{\tau}$ (left column) to friction coefficient $\mu$ (right column). See \ref{['sec:tradyn:terrainlayouts']} for details.
• Figure 5: Prediction error evaluation for the proposed model and its ablations (no terrain lookup / no calibration), plotted over the prediction horizon (number of prediction steps). From left to right: Positional error (euclidean distance), velocity error (absolute difference), angular error (absolute difference). Depicted are the mean and 20%, 80% percentiles over 150 evaluation rollouts for 5 independently trained models per model variant. Our approach with terrain lookup and calibration clearly outperforms the other variants in position and velocity prediction (left and center panel). For predicting the angle (right panel), terrain friction is not relevant, which is why the terrain lookup brings no advantage. However, calibration is important for accurate angle prediction. See \ref{['sec:tradyn:eval_prediction']} for details.