Abstract:Trajectory planning is a fundamental problem in robotics. It facilitates a wide range of applications in navigation and motion planning, control, and multi-agent coordination. Trajectory planning is a difficult problem due to its computational complexity and real-world environment complexity with uncertainty, non-linearity, and real-time requirements. The multi-agent trajectory planning problem adds another dimension of difficulty due to inter-agent interaction. Existing solutions are either search-based or optimization-based approaches with simplified assumptions of environment, limited planning speed, and limited scalability in the number of agents. In this work, we make the first attempt to reformulate single agent and multi-agent trajectory planning problem as query problems over an implicit neural representation of trajectories. We formulate such implicit representation as Neural Trajectory Models (NTM) which can be queried to generate nearly optimal trajectory in complex environments. We conduct experiments in simulation environments and demonstrate that NTM can solve single-agent and multi-agent trajectory planning problems. In the experiments, NTMs achieve (1) sub-millisecond panning time using GPUs, (2) almost avoiding all environment collision, (3) almost avoiding all inter-agent collision, and (4) generating almost shortest paths. We also demonstrate that the same NTM framework can also be used for trajectories correction and multi-trajectory conflict resolution refining low quality and conflicting multi-agent trajectories into nearly optimal solutions efficiently. (Open source code will be available at

What problem does this paper attempt to address?

The paper attempts to address the issues of computational complexity and environmental uncertainty in trajectory planning, particularly in single-robot and multi-robot systems. Specifically: 1. **Computational Complexity**: Existing search algorithms or optimization methods take a long time to compute when handling single-robot and multi-robot trajectory planning due to the need for repeated searches for each instance. This is especially true in multi-robot scenarios, where the computational complexity increases sharply with the number of robots. 2. **Environmental Complexity**: Real-world environments typically have uncertainty, non-linearity, and real-time requirements, making trajectory planning more difficult. Existing methods often rely on simplified environmental assumptions, resulting in limited planning speed and poor scalability. 3. **Multi-Robot Interaction**: Multi-robot trajectory planning not only needs to consider the path of each robot but also avoid collisions between robots, which increases the difficulty of the problem. To address these issues, the paper proposes a new method—the Neural Trajectory Model (NTM), which reformulates the single-robot and multi-robot trajectory planning problem as a query problem of implicit neural representations. Through this approach, NTM can quickly generate near-optimal trajectories in complex environments and demonstrates the following advantages in experiments: - **Sub-millisecond Planning Time**: Achieves very fast planning speed using GPU. - **Almost All Environment Collisions Avoided**: Ensures that the generated trajectories do not collide with obstacles in the environment. - **Almost All Robot Collisions Avoided**: Ensures that multiple robots do not collide with each other. - **Generates Near-Shortest Paths**: The length of the generated trajectories is close to the optimal solution. Additionally, the paper demonstrates the potential of NTM in trajectory correction and multi-trajectory conflict resolution, efficiently optimizing low-quality or highly conflicting trajectories into near-optimal solutions.

Neural Trajectory Model: Implicit Neural Trajectory Representation for Trajectories Generation

Adaptive Model Prediction Control-Based Multi-Terrain Trajectory Tracking Framework for Mobile Spherical Robots

Multi-Agent Trajectory Prediction Based on Graph Neural Network

Efficient Optimization-Based Trajectory Planning for Unmanned Systems in Confined Environments

A Model Predictive Trajectory Planning Framework for Autonomous Ground Vehicles on Structured and Unstructured Roads.

Learning to Plan Maneuverable and Agile Flight Trajectory with Optimization Embedded Networks

Leveraging Neural Network Gradients within Trajectory Optimization for Proactive Human-Robot Interactions

TrajAgent: An Agent Framework for Unified Trajectory Modelling

Trajectory Generation in New Environments from Past Experiences

Modeling Trajectories with Neural Ordinary Differential Equations

An Efficient Spatial-Temporal Trajectory Planner for Autonomous Vehicles in Unstructured Environments

Efficient Trajectory Generation in 3D Environments with Multi-Level Map Construction

Trajectory Forecasts in Unknown Environments Conditioned on Grid-Based Plans

Physics-informed Neural Mapping and Motion Planning in Unknown Environments

Spatio-Temporal-Interaction Graph Neural Networks for Multi-Agent Trajectory Prediction

OCTNet: Trajectory Generation in New Environments from Past Experiences

Parametrization Approach for Real-Time Generation of Minimum-Effort Trajectories via Neural Network

Reachability-based Trajectory Design with Neural Implicit Safety Constraints

Reachability-based Trajectory Design via Exact Formulation of Implicit Neural Signed Distance Functions

A Novel Resolution Scheme of Time-Energy Optimal Trajectory for Precise Acceleration Controlled Industrial Robot Using Neural Networks

Dynamic Neural Potential Field: Online Trajectory Optimization in Presence of Moving Obstacles