Learning Approximated Maximal Safe Sets via Hypernetworks for MPC-Based Local Motion Planning

Bojan Derajić,Mohamed-Khalil Bouzidi,Sebastian Bernhard,Wolfgang Hönig
2024-10-27
Abstract:This paper presents a novel learning-based approach for online estimation of maximal safe sets for local motion planning tasks in mobile robotics. We leverage the idea of hypernetworks to achieve good generalization properties and real-time performance simultaneously. As the source of supervision, we employ the Hamilton-Jacobi (HJ) reachability analysis, allowing us to consider general nonlinear dynamics and arbitrary constraints. We integrate our model into a model predictive control (MPC) local planner as a safety constraint and compare the performance with relevant baselines in realistic 3D simulations for different environments and robot dynamics. The results show the advantages of our approach in terms of a significantly higher success rate: 2 to 18 percent over the best baseline, while achieving real-time performance.
Robotics,Machine Learning,Systems and Control
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: in the local motion planning tasks of mobile robots, how to estimate maximal safe sets in real - time to ensure the safety and performance of robots when performing tasks in complex and unknown environments. Specifically, the researchers proposed a new learning method based on hypernetworks, aiming to estimate the maximal safe sets online through local observations and integrate them as constraints into the model - predictive - control (MPC) local planner, thereby achieving a higher success rate and real - time performance. ### Problem Background 1. **Safety Challenges** - In the field of autonomous robots, especially when operating in natural unstructured and uncertain environments, ensuring the safety of robots is a key challenge. - Modern robots need to maintain strict safety standards in environments shared with humans while also having efficient operating performance. 2. **Limitations of Existing Methods** - Although traditional optimization, sampling, and learning - based algorithms perform well in some aspects, they often face challenges in computational complexity and real - time performance when dealing with complex nonlinear dynamic systems and high - dimensional environments. - Although methods based on Hamilton - Jacobi (HJ) reachability analysis can provide strict mathematical guarantees, their computational complexity is high, making it difficult to achieve real - time applications. ### Core Contributions of the Paper 1. **Real - Time Estimation of Maximal Safe Sets** - A learning method based on hypernetworks was proposed, which can estimate maximal safe sets in real - time according to local observations in unknown environments. - By combining the hypernetwork with the main network, the parameterization of different binary cost maps was achieved, enabling the main network to quickly generate corresponding maximal safe set estimates in real - time environments. 2. **Integration into MPC Local Planner** - The estimated maximal safe sets were integrated as constraints into the MPC local planner, ensuring safety during the planning process. - Experimental results show that this method significantly improves the success rate of tasks in different environments and robot dynamics, with an improvement of 2% to 18% compared to the benchmark method. 3. **Improved Real - Time Performance** - Compared to existing methods based on HJ reachability, this method can achieve real - time performance in continuously changing high - dimensional local observations and is suitable for time - critical systems. ### Technical Details - **System Model** \[ \dot{x}(t)=f(x(t), u(t)) \] where \(x(t)\in\mathbb{R}^n\) is the state vector, \(u(t)\in\mathbb{R}^m\) is the control vector, and \(t\in\mathbb{R}\) is the time variable. - **MPC Optimal Control Problem** \[ \min_{u_k:k + N - 1}p(x_{k + N})+\sum_{i = 0}^{N - 1}q(x_{k + i}, u_{k + i}) \] Constraints include: \[ x_{k + i+1}=f_d(x_{k + i}, u_{k + i}),\quad i = 0:N - 1 \] \[ x_{k + i}\in X,\quad i = 0:N \] \[ u_{k + i}\in U,\quad i = 0:N - 1 \] \[ h_k(x_{k + i})\geq0,\quad i = 0:N \] - **Value Function** \[ V(x, t)=\sup_{u(\cdot)}\left\{J(x, t, u(\cdot))\right\} \] where \(J(x, t, u(\cdot))=\min_{\tau\in[t, T]}l(\xi^u_{x,t}(\tau))\), \(l(x)\) is the objective function, and \(\xi^u_{x,t}(\tau)\) is the system