Generating Freeform Endoskeletal Robots

Muhan Li,Lingji Kong,Sam Kriegman

2024-12-02

Abstract:The automatic design of embodied agents (e.g. robots) has existed for 31 years and is experiencing a renaissance of interest in the literature. To date however, the field has remained narrowly focused on two kinds of anatomically simple robots: (1) fully rigid, jointed bodies; and (2) fully soft, jointless bodies. Here we bridge these two extremes with the open ended creation of terrestrial endoskeletal robots: deformable soft bodies that leverage jointed internal skeletons to move efficiently across land. Simultaneous de novo generation of external and internal structures is achieved by (i) modeling 3D endoskeletal body plans as integrated collections of elastic and rigid cells that directly attach to form soft tissues anchored to compound rigid bodies; (ii) encoding these discrete mechanical subsystems into a continuous yet coherent latent embedding; (iii) optimizing the sensorimotor coordination of each decoded design using model-free reinforcement learning; and (iv) navigating this smooth yet highly non-convex latent manifold using evolutionary strategies. This yields an endless stream of novel species of "higher robots" that, like all higher animals, harness the mechanical advantages of both elastic tissues and skeletal levers for terrestrial travel. It also provides a plug-and-play experimental platform for benchmarking evolutionary design and representation learning algorithms in complex hierarchical embodied systems.

Robotics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to automatically design a new type of endoskeletal soft robots, which can combine the advantages of rigid and flexible structures to achieve efficient movement on complex terrains. Specifically, the paper mainly solves the following problems:

1. **Limitations of existing robot designs**: Traditional robot designs are usually limited to two extreme types - fully rigid jointed bodies and fully soft joint - less bodies. These designs have very limited morphological assumptions and cannot fully utilize the mechanical advantages of elastic tissues and skeletal levers.

2. **Challenges in automatic design**: The process of automatically designing robots is very complex because the objectives and variables of controller optimization depend on the optimization of discrete mechanical structures (such as the distribution of motors and limbs). This makes automatic design extremely difficult.

3. **Structural optimization in multi - physics simulation**: In order to generate external and internal structures simultaneously, a multi - physics simulator that can handle the two - way coupling of rigid and flexible bodies is required, and this simulator must be able to support structural optimization.

To solve these problems, the author proposes a new method, which is achieved through the following steps:

- **Modeling and coding**: Use a 3D end - to - end design strategy to model the endoskeletal robot as a collection of integrated elastic cells and rigid cells, which are directly connected to form soft tissues and composite rigid bodies.
- **Continuous latent embedding**: Encode these discrete mechanical subsystems into continuous but coherent latent embeddings.
- **Reinforcement learning optimization**: Use model - free reinforcement learning to optimize the sensor - motor coordination of each decoded design.
- **Evolutionary strategy navigation**: Use evolutionary strategies to navigate in the smooth but highly non - convex latent manifold to generate a series of novel "high - level robots".

Through this method, the author successfully created an experimental platform for benchmarking the application of evolutionary design and representation learning algorithms in complex hierarchical physical systems. Eventually, an endoskeletal soft robot that can move efficiently in multiple environments was achieved.

In terms of formulas, the formulas involved in the paper are mainly used to describe the dynamic equations and reward functions in the simulator, for example:
\[ Mv\ddot{X}_v = F_v^b + F_v^{ext} + I(v \in bone)F_v^c \]
where \( Mv \) is the generalized mass matrix, \( \ddot{X}_v \) is the second - order derivative of the generalized position with respect to time, \( F_v^b \) and \( F_v^{ext} \) are the internal and external force terms respectively, and \( F_v^c \) is the constraint force term.

The reward function is defined as:
\[ r_t = \begin{cases}
-Plarge, & \text{if } \|v_t\| < \delta \\
\max(\hat{u}_t \cdot v_t, -Psmall), & \text{otherwise}
\end{cases} \]

These formulas ensure that the behavior and performance evaluation of the robot in the simulation have a scientific basis and interpretability.