The field of robotics has seen significant advancements in recent years, with companies like Boston Dynamics pushing the limits of what robots can achieve. One key concept enabling robots to perform highly dynamic, coordinated movements is sequential composition. This approach involves combining multiple behaviors or actions in a specific sequence to accomplish a job.

Sequential composition is a fundamental concept in control systems, robotics, and motion planning that enables the development of advanced robotic systems capable of performing complex tasks in dynamic and unstructured environments. At its core, sequential composition refers to the process of designing, implementing, and executing a series of behaviors or actions in a predefined or adaptive order to achieve a specific objective. This approach provides robots with the ability to switch between different behaviors and actions based on the current situation, task requirements, or environmental conditions.

Sequential composition in athletic mobility likely refers to the development of algorithms and control systems that enable robots to perform a series of movements or tasks in a coordinated and efficient manner. These movements are designed based on principles of biomechanics, kinematic and dynamic optimization, and trajectory optimization. This enables the robots to exhibit human-like or animal-like behaviors and adapt to unstructured and dynamic environments.

By combining these principles with innovations in coupling mechanics, software and hardware control units, and real-time correlation models, robotic systems like those of Boston Dynamics can achieve a high level of mobility and adaptability, allowing them to perform tasks that were previously difficult or impossible for robots to accomplish.

Let’s consider the example of the Boston Dynamics’ robot dog, Spot. Spot is designed to navigate and interact with various environments, which requires a combination of multiple behaviors and actions. Sequential composition plays a crucial role in Spot’s ability to perform complex tasks, such as navigating uneven terrain, climbing stairs, etc. By coordinating different behaviors and actions, Spot’s control system can adapt its movements in real-time, allowing it to successfully complete these tasks.

Imagine Spot is tasked with navigating through an obstacle course consisting of uneven terrain, stairs, and a door that needs to be opened. In this case, Spot’s control system would use sequential composition to coordinate its actions and successfully complete the course.

  • Uneven terrain: Spot would first utilize its balance and stability algorithms to maintain its posture while walking across the uneven surface. This could involve adjusting the legs’ positions and angles to maintain stability while moving forward.
  • Stairs: Once it reaches the stairs, Spot would switch to a different set of behaviors and control algorithms specifically designed for stair climbing. This could involve adjusting the leg movements, step height, and timing to ascend or descend the stairs safely.
  • Door opening: Upon reaching the door, Spot would employ another set of behaviors and actions to manipulate the door handle and push or pull the door open. This might involve using its robotic arm, adjusting its body position, and applying the necessary force to open the door.

In each of these steps, Spot combines different behaviors or actions in a specific sequence to achieve the complex task of navigating the obstacle course. The control system coordinates these actions by switching between various algorithms and motion planning strategies to ensure seamless transitions and fluid movement. This is an example of how sequential composition is applied in robotics to perform highly dynamic, coordinated movements.

Core Steps in Sequential Composition for Robotics

The process of sequential composition in robotics typically involves the following components:

  • Behavior or action representation: Individual behaviors or actions are represented using mathematical models, such as differential equations or state-space models, which describe the underlying dynamics and kinematics of the robot during the execution of a specific behavior.
  • High-level planning: A high-level planner or decision-making module generates a sequence of behaviors or actions that the robot should execute to accomplish a given task. This could be based on heuristic algorithms, optimization techniques, or learning-based methods, depending on the specific problem and robotic system.
  • Low-level control: For each behavior or action in the sequence, a low-level controller is designed to ensure the proper execution of the behavior, considering the robot’s physical constraints, actuator limitations, and environmental interactions. This may involve the use of classical control techniques, such as PID control, model predictive control, or advanced methods like adaptive control and robust control.
  • Transition management: To ensure seamless transitions between consecutive behaviors or actions, transition management strategies are employed. These strategies can involve smooth interpolation between the end state of one behavior and the initial state of the subsequent behavior, or the design of hybrid controllers that can handle the switching between different control modes while maintaining stability and performance.
  • Adaptation and learning: Sequential composition can also incorporate adaptation and learning mechanisms, which enable the robot to adjust its behavior or action sequence based on the current situation, performance feedback, or new information about the environment. This can involve the use of reinforcement learning, supervised learning, or unsupervised learning techniques.

The concept of sequential composition has practical implications across various industries and applications. Some examples include:

    • Search and rescue operations: Robots with advanced mobility can navigate disaster zones and assist in locating survivors.
    • Construction and inspection: Robots can access hard-to-reach areas and perform tasks that would be hazardous for humans.
    • Agriculture: Robots can monitor and tend to crops, efficiently performing tasks like harvesting, pruning, and planting.
    • Healthcare and assisted living: Robots can aid in physical therapy and provide assistance to the elderly or disabled.

    Sequential composition in robotics is closely related to several theoretical concepts in control systems, motion planning, and artificial intelligence. Here are some of the key theoretical details:

    • Hybrid Systems and Hybrid Control: Hybrid systems are those that exhibit both continuous and discrete dynamics. In robotics, a hybrid system can represent the combination of continuous motions (e.g., walking or grasping) and discrete events (e.g., switching between behaviors). Hybrid control theory is a branch of control systems that deals with the design of controllers for hybrid systems. Sequential composition in robotics often leverages hybrid control to switch between different behaviors or actions while maintaining stability and performance.
    • Finite State Machines (FSMs): FSMs are a mathematical model of computation that can be used to represent and design the control logic for sequential composition. An FSM consists of a set of states, transitions between those states, and actions performed in each state. In the context of robotics, states can represent different behaviors or actions, while transitions are triggered by specific conditions or events. FSMs can be used to design the high-level control logic for sequential composition, dictating when and how a robot should switch between different behaviors.
    • Motion Planning and Trajectory Optimization: Motion planning is the process of finding feasible paths for a robot to achieve a specific goal, while trajectory optimization deals with determining the optimal path considering constraints and objectives. In sequential composition, motion planning algorithms are employed to generate a sequence of desired behaviors or actions, and trajectory optimization techniques are used to compute the optimal way to execute those actions. These methods often incorporate concepts from optimization theory, graph theory, and sampling-based planning.
    • Reinforcement Learning (RL): RL is a type of machine learning where an agent learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or penalties. In robotics, RL can be used to learn and optimize sequential composition policies. The robot can learn to switch between different behaviors or actions based on the environmental conditions and task requirements. This allows the robot to adapt to new situations and improve its performance over time.
    • Hierarchical Control: Hierarchical control is a control architecture that organizes the control system into multiple levels, with each level responsible for a different aspect of the robot’s behavior. Sequential composition can be implemented within a hierarchical control framework, where high-level control focuses on deciding which behaviors or actions to execute, while low-level control deals with the actual execution of those actions. This separation of concerns allows for better modularity and adaptability in robotic systems.

    Other Examples

    Case study 1: Let’s consider a simple example of a mobile robot with a robotic arm that needs to perform a pick-and-place task in a warehouse environment. The robot’s objective is to pick up an object from one location and place it at another location while navigating around obstacles. Sequential composition can be used to break down this complex task into a series of simpler behaviors or actions:

    • Navigate to the object: The robot first uses its motion planning algorithms to find a collision-free path from its current position to the location of the object. It then follows this path using its wheeled or legged locomotion system.

    • Align with the object: Upon reaching the object’s location, the robot adjusts its position and orientation so that its robotic arm is correctly aligned with the object. This may involve using visual feedback from cameras or other sensors to fine-tune the alignment.

    • Grasp the object: The robot then executes a grasping behavior, which involves controlling its robotic arm and gripper to pick up the object securely. This may require precise control of the arm’s joints and the application of appropriate gripping forces.

    • Navigate to the destination: With the object securely grasped, the robot plans and follows a new collision-free path from its current position to the destination location, where it needs to place the object.

    • Place the object: Upon reaching the destination, the robot positions its robotic arm to release the object at the desired location. This may involve controlling the arm’s joints and gripper to place the object gently and accurately.

    In this example, the robot performs a sequence of five distinct behaviors or actions—navigation, alignment, grasping, navigation, and placement—using sequential composition. The robot’s control system coordinates these actions and manages the transitions between them to ensure fluid movement and successful completion of the pick-and-place task.

    Case study 2: Let’s consider a different example, this time with a humanoid robot that needs to perform a task in a domestic setting, such as preparing a simple meal. Sequential composition can be used to break down this complex task into a series of simpler behaviors or actions:

    • Navigate to the refrigerator: The humanoid robot first uses its motion planning algorithms to find a collision-free path from its current position to the refrigerator. It then walks along this path, adjusting its leg movements and balance as needed.

    • Open the refrigerator door: Upon reaching the refrigerator, the robot uses its arm and gripper to grasp the door handle and pull the door open. This requires precise control of the arm’s joints and the application of appropriate forces to open the door smoothly.

    • Retrieve ingredients: The robot identifies the necessary ingredients inside the refrigerator using its cameras and object recognition algorithms. It then carefully grasps each ingredient with its gripper and removes it from the refrigerator.

    • Close the refrigerator door: With the ingredients in hand, the robot uses its free arm to close the refrigerator door by pushing it gently until it is fully closed.

    • Navigate to the kitchen counter: Holding the ingredients, the robot plans and follows a new collision-free path from its current position to the kitchen counter, adjusting its leg movements and balance as needed.

    • Place ingredients on the counter: Upon reaching the counter, the robot carefully places the ingredients on the surface, ensuring they are properly positioned for the next step.

    • Prepare the meal: The robot uses its arms, grippers, and various kitchen tools (such as a knife or a spoon) to perform the required actions to prepare the meal, like chopping vegetables or stirring a sauce. This may involve a series of fine motor skills and precise control of the arm’s joints and grippers.

    • Serve the meal: Once the meal is prepared, the robot plates the food and navigates to the dining table to serve it. It plans and follows another collision-free path, ensuring it maintains balance while carrying the plated meal.

    In this example, the humanoid robot performs a sequence of eight distinct behaviors or actions using sequential composition. The robot’s control system coordinates these actions and manages the transitions between them, ensuring fluid movement and successful completion of the meal preparation task.

    Case study 3: We have now an example of a robotic vacuum cleaner that needs to clean a room while avoiding obstacles and returning to its charging station after completing the task. Sequential composition can be used to break down this task into a series of simpler behaviors or actions:

    • Start cleaning: The robotic vacuum cleaner begins cleaning the room using its predefined cleaning pattern, such as a spiral or grid pattern, to cover the entire floor area efficiently.

    • Obstacle detection: As the robot moves around the room, it uses its sensors (e.g., infrared, ultrasonic, or LIDAR) to detect obstacles, such as furniture or objects on the floor.

    • Obstacle avoidance: Upon detecting an obstacle, the robot adjusts its cleaning pattern to avoid the obstacle, navigating around it while still maintaining coverage of the floor area. This may involve changing its direction, speed, or temporarily altering the cleaning pattern.

    • Resume cleaning: Once the obstacle has been successfully avoided, the robot resumes its predefined cleaning pattern, ensuring that it continues to cover the entire floor area.

    • Monitor battery level: Throughout the cleaning process, the robot monitors its battery level. When the battery level drops below a certain threshold, the robot determines that it’s time to return to the charging station.

    • Navigate to the charging station: The robot uses its mapping and localization algorithms to find the shortest and safest path back to its charging station. It then follows this path, avoiding any obstacles encountered along the way.

    • Dock with the charging station: Upon reaching the charging station, the robot aligns itself with the station and connects to the charging contacts. This may involve using sensors or computer vision algorithms to accurately position itself for docking.

    In this example, the robotic vacuum cleaner performs a sequence of seven distinct behaviors or actions using sequential composition. The robot’s control system coordinates these actions and manages the transitions between them, ensuring fluid movement, efficient cleaning, and successful navigation back to the charging station when needed.


    Sequential composition is a transformative concept in robotics, providing a structured and adaptable framework for integrating multiple behaviors or actions to achieve complex tasks. As robotic systems continue to evolve, the importance of sequential composition will only increase, enabling robots to seamlessly operate across a wider range of applications and environments.

    As artificial intelligence and machine learning techniques advance, we can envision a future where sequential composition is further enhanced by self-learning and adaptation capabilities. Robots will not only be able to execute predefined sequences of behaviors, but also dynamically generate and optimize their action sequences based on real-time feedback and environmental changes. This will lead to even more versatile, resilient, and efficient robotic systems, capable of tackling challenges that are currently beyond our reach.

    Furthermore, the fusion of advanced sensing, communication, and control technologies will enable more sophisticated implementations of sequential composition, allowing multiple robots to work together in a coordinated manner. Collaborative robotic systems will be able to share information, synchronize their actions, and adapt their behavior sequences in real-time, leading to unprecedented levels of teamwork and efficiency in various domains.

    Ultimately, the ongoing development of sequential composition will pave the way for a new era of robotics, characterized by increased autonomy, adaptability, and collaboration. The impact of these advancements will be felt across numerous industries and applications, from manufacturing and agriculture to healthcare and disaster relief, significantly improving our quality of life and shaping the future of human-robot interaction.


    Credits: Stelian Brad