Skip to main content

Chapter 12: Physics Simulation in Gazebo

A key feature that sets Gazebo apart from other 3D visualizers is its ability to simulate physics. This allows you to create a dynamic digital twin that responds to forces and interacts with its environment in a physically plausible way. This chapter dives into the core concepts of physics simulation in Gazebo.

The Role of Physics Engines​

Gazebo does not have its own built-in physics engine. Instead, it uses a plugin-based architecture to interface with existing open-source physics engines. The most common ones are:

  • ODE (Open Dynamics Engine): The default engine in many older versions of Gazebo. It's fast and well-suited for many robotics applications.
  • Bullet: A popular engine known for its performance and use in games and movies.
  • Simbody: An engine designed for high-fidelity simulation of articulated biomechanical and mechanical systems, like humanoid robots.
  • DART: An engine that excels at simulating systems with many contacts and closed-loop kinematics.

You can choose the physics engine that best suits your simulation needs, and you can configure its properties in the world SDF file.

Rigid Body Dynamics​

In Gazebo, every object with physical properties is treated as a rigid body. This means it has properties like:

  • Mass: The amount of matter in the object.
  • Inertia: A tensor that describes the object's resistance to rotational motion.
  • Gravity: A global force that pulls objects downwards.

The physics engine uses these properties to calculate the object's motion when forces are applied to it. For a robot, each link (the rigid parts between the joints) is a separate rigid body.

Collision Detection and Contact Simulation​

For a simulation to be useful, objects can't pass through each other. Collision detection is the process of figuring out when two rigid bodies are intersecting. In Gazebo, this is done using collision geometries, which are simplified shapes (like spheres, boxes, or cylinders) that approximate the visual shape of the object.

When a collision is detected, the physics engine simulates a contact. This involves calculating contact forces to prevent penetration and simulating friction. Key concepts include:

  • Friction: The force that resists motion between two surfaces in contact. Gazebo models both static friction (when at rest) and dynamic friction (when moving).
  • Restitution (Bounciness): The amount of energy an object retains after a collision. A value of 1.0 means it's perfectly elastic (like a bouncy ball), while 0.0 means it's perfectly inelastic (like a lump of clay).

Joint Dynamics​

In addition to the dynamics of individual links, Gazebo also simulates the dynamics of the joints that connect them. This includes:

  • Joint Limits: The minimum and maximum angle or position a joint can reach.
  • Damping: A dissipative force that acts to slow down the joint's motion, similar to air resistance or hydraulic damping. This is crucial for creating stable simulations.
  • Friction: An internal friction within the joint that resists motion.

Tuning for Realism and Stability​

A common challenge in physics simulation is that the default parameters are often not stable or realistic for a specific robot. Tuning is the process of adjusting physics parameters to achieve the desired behavior. This often involves a trade-off between realism and stability.

Some common parameters to tune include:

  • Time Step: The amount of time the physics engine advances in each simulation step. A smaller time step can increase accuracy and stability but requires more computation.
  • Solver Iterations: The number of times the physics solver runs per time step. More iterations can improve the accuracy of contact and joint constraints.
  • Contact Stiffness and Damping (CFM and ERP): Advanced parameters that can help stabilize contacts and prevent objects from "exploding" or becoming jittery.

Achieving a high-fidelity digital twin often requires a careful process of tuning these parameters until the simulated robot's behavior closely matches the behavior of the real robot.