Trajectory optimization for a class of robots belonging to Constrained Collaborative Mobile Agents (CCMA) family

Nitish Kumar, Stelian CorosICRA 2020 · DOI

TL;DR

This paper introduces a novel class of robots called Constrained Collaborative Mobile Agents (CCMAs), which combine non-holonomic mobile bases with active-passive closed-loop kinematic chains. They also present a flexible optimization framework that allows for simultaneous optimization of both the robot's design parameters and its motion trajectory, enabling highly adaptable robotic systems.

What Problem Does This Paper Address?

The paper tackles the challenge of designing and controlling complex robotic systems that require both mobility and manipulation capabilities. Traditional approaches often treat robots as monolithic entities designed for a specific task. However, the authors envision a more modular and reconfigurable robotic ecosystem. They focus on a specific type of modular system: multiple mobile bases connected by kinematic chains (like arms). These systems, termed CCMAs, present a significant challenge because the non-holonomic constraints of the mobile bases, combined with the kinematics of the chains, make trajectory planning and optimization non-trivial. Furthermore, optimizing only the motion without considering how the robot's physical design affects its capabilities limits its adaptability.

Methods

The core contribution is a trajectory optimization framework that can handle CCMAs. This framework is designed to be general, working with varying numbers of mobile robots, different kinematic chain topologies, and various actuator placements.

Key methodological aspects include:

  • Modeling: The system's rigid body kinematics are modeled using a constraint-based formulation, abstracting the system into interconnected rigid bodies.
  • Trajectory Optimization: The paper employs a gradient-based optimization approach. The objective function is designed to achieve specific goals (like end-effector positioning) while also incorporating constraints for collision avoidance (internal and external) and smooth control inputs.
  • Concurrent Design and Control Optimization: A novel aspect is the ability to optimize both the robot's design parameters (e.g., link lengths, geometry) and its control policy (trajectory) simultaneously. This is achieved by including design parameters in the optimization problem alongside control inputs.
  • Sensitivity Analysis: To efficiently compute the gradients required for optimization, the authors utilize sensitivity analysis techniques to analytically derive how changes in control inputs and design parameters affect the system's state and objective function.
  • Forward Simulation: The optimization process relies on a forward simulation step to predict the system's state transitions based on control inputs and design parameters.

Datasets

Not discussed in the paper. The paper focuses on simulation and experimental validation of their proposed method using custom-designed CCMA prototypes.

Key Insights and Ideas

  • Modularity and Reconfigurability: The CCMA concept itself is a significant idea, moving towards a future of highly adaptable and reconfigurable robotic systems rather than fixed, single-purpose robots.
  • Concurrent Optimization is Powerful: Optimizing robot design alongside its motion planning is crucial for maximizing performance and adaptability, especially in early design stages or for evolving tasks. This allows the robot to "grow into" its task.
  • Generality of the Framework: The optimization framework is designed to be independent of the specific CCMA configuration (number of robots, DOF, morphology), making it broadly applicable to this class of systems.
  • Handling Non-holonomic Constraints: The method effectively addresses the inherent difficulties of planning for non-holonomic mobile bases, which cannot move sideways directly.

Key Results

The paper demonstrates the effectiveness of their framework through simulations and experiments:

  • Standalone Trajectory Optimization: They successfully optimized trajectories for tasks like end-effector positioning, internal mobile agent collision avoidance, and external obstacle avoidance.
  • Concurrent Design and Control Optimization: They showed that this approach can improve the CCMA system's workspace and manipulation capabilities by adapting the design to the task. Figure 9 illustrates this, showing how concurrent optimization leads to a more effective trajectory.
  • Physical Prototype Validation: Experimental results with two 4-DOF prototypes (tracked mobile bases) validated the simulation findings, showing good agreement between simulated and captured motion data. RMSE values for end-effector pose were reported to be within acceptable ranges (e.g., around 0.025-0.072 m for position, 0.04-0.062 rad for orientation).

Limitations and Future Work

  • Initialization: The paper notes that the initialization of control inputs can be trivial (all zeros) but suggests that better initialization strategies (e.g., from previous solutions or RRT) could improve convergence and the final solution.
  • Mobile Base Orientation: The current framework assumes trivial initial orientations for mobile bases. Allowing for arbitrary starting orientations would expand the feasible solution space.
  • Heterogeneous Systems: While the framework is general, future work aims to explore motion generation with a more heterogeneous mix of mobile robots (omni-directional, non-holonomic, quadruped, aerial).
  • Force Transmission: The current focus is on motion generation. Future work will also explore optimizing force transmission capabilities concurrently with motion.

Summary updated

BibTeX
@inproceedings{kumar2020trajectory,
  author = {Nitish Kumar and Stelian Coros},
  title = {{Trajectory optimization for a class of robots belonging to Constrained Collaborative Mobile Agents (CCMA) family}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2020},
  doi = {10.1109/icra40945.2020.9197048},
}
Figure