Probabilistic Road Map (PRM)

A probabilistic roadmap (PRM) is a motion planning algorithm used in robotics to find a

collision-free path between a robot’s starting and goal configurations. Here’s how it works:
1. Construction Phase:
o Random samples are taken from the robot’s configuration space.
o These samples are tested to determine if they lie in free space (i.e., not in collision with
obstacles).
o Local planners connect nearby configurations to create a graph (roadmap).
o The roadmap is built until it becomes dense enough.
2. Query Phase:
o The start and goal configurations are connected to the roadmap.
o A shortest path is found using Dijkstra’s algorithm.
3. Completeness and Convergence:
o PRM is probabilistically complete, meaning that as the number of samples increases, the
probability of finding a path (if one exists) approaches 1.
o The rate of convergence depends on visibility properties of the free space.

PRM Sampling Stragies

sampling strategies in probabilistic roadmap (PRM) motion planning, there are several
approaches to consider:
1. Random Sampling:
o This method involves randomly sampling the entire configuration space (C-space).
o However, it’s important to note that collision-free areas are easier to navigate and don’t require
as many samples.
o The focus should be on collision regions, where the planner needs denser samples for tight
navigation areas.
2. Obstacle-Based PRM (OBPRM):
o In OBPRM, if a configuration q is in collision, the planner re-samples in the vicinity of the
collision to find a safe configuration near the obstacle.
o It chooses a random direction and generates a nearby sample in the free space.
o This approach biases sampling toward regions where collisions are likely

PRM Connection Stragies

probabilistic roadmap (PRM) motion planning, the connection strategy plays a crucial role in
constructing the roadmap. Let’s delve into some connection strategies:
1. Slicing Connection Strategy:
o This innovative approach addresses the challenge of dealing with high-dimensional configuration
spaces (C-spaces).
o During roadmap construction, it identifies components of motion and plans roadmap edges on a
sequence of configuration slices that approximate the C-spaces of these components.
o By doing so, it extends the dimensionality of problems PRM planners can handle.
o Experimental results show success in solving problems involving serpentine robots with up to
150 degrees-of-freedom and multi-robot systems with up to 144 degrees-of-freedom1.
2. Coarse Connection:
o In this strategy, the roadmap’s connected components are maintained.
o It attempts to connect two milestones only if they belong to distinct components.
o Coarse connection helps address the “narrow passage” limitation of PRMs and improves
connectivity in the roadmap

Single based sampling based Planner

Single-query sampling-based planners are efficient algorithms used to quickly solve motion
planning problems. Let’s explore the key aspects of these planners:
1. Objective:
o Given a robotic system with constraints, the goal is to find a continuous valid path from a
specified start state to a goal state.
2. Common Core:
o Single-query planners recompute the roadmap for each query.
o They handle environments where conditions can change across queries (but not within a single
query).
3. Rapidly-Exploring Random Trees (RRTs):
o RRTs are a popular single-query planner.
o They grow a tree by randomly sampling configurations and connecting them.
o When inside a pre-specified goal region, the search stops.
o Path retrieval is done by retaining pointers to parents in the tree.
4. Completeness and Optimality:
o RRTs are probabilistically complete, meaning they find a solution if one exists.
o They aim for suboptimal paths quickly and improve over time.
o Strategies include keeping old trees or building new ones from scratch.
5. Improving RRT Paths:
o Simple Shortcutting Algorithm:
 Repeatedly select two points randomly and attempt to connect them.
 If the resulting path is shorter than the previous one, replace it.

Rapidly exploring random trees

rapidly exploring random tree (RRT) is an algorithm designed to efficiently search

nonconvex, high-dimensional spaces by randomly building a space-filling tree. Let’s break down
its key features:
1. Construction Process:
o The RRT grows a tree incrementally from randomly sampled configurations in the search space.
o Each sample attempts to connect to the nearest state in the existing tree.
o If the connection is feasible (i.e., passes through free space and obeys constraints), the new state
is added to the tree.
2. Space-Filling Bias:
o RRTs preferentially expand towards large unsearched areas.
o The probability of expanding an existing state is proportional to the size of its Voronoi region.
o Voronoi regions of frontier states (unexplored areas) receive more attention.
3. Growth Factor:
o The length of connections between the tree and new states is limited by a growth factor.
o If a random sample is too far from the nearest tree state, a new state along the line to the sample
is used instead.
o This balances incremental growth while maintaining the space-filling bias.
4. Guided Search:
o Practical RRT implementations bias sampling towards problem goals.
o Introducing a small probability of sampling the goal helps guide the search.
5. Applications:
o RRTs are widely used in autonomous robotic motion planning.
o They handle obstacles and differential constraints (nonholonomic and kinodynamic).
o RRTs generate open-loop trajectories for nonlinear systems with state constraints.

SBL Planner

SBL (Single-Query Bi-directional Lazy) planner:

1. SBL Planner Overview:
o The SBL planner is designed for single-query motion planning.
o It builds two trees concurrently: one from the initial configuration and another from the goal
configuration.
o The goal is to iteratively connect milestones in these trees until a collision-free path is found.
2. Milestone Sampling Strategies:
o SBL employs milestone sampling strategies to efficiently explore the configuration space.
o Multi-stage sampling:
 Uniformly generate milestones and paths.
 Enhance by selecting more milestones around narrow areas (e.g., Gaussian sampling).
o Obstacle-sensitive sampling:
 Retain configurations as milestones only if collision-free and forbidden configurations are
neighbors.
 Helps capture boundaries of narrow passages.
 First roadmap dilates the free space to widen narrow passages for easier connections.
3. Connection Strategies:
o Delayed Collision-Checking:
 Collision checking consumes most of the runtime.
 Avoid collision tests until absolutely necessary.
o Lazy Collision-Checking:
 Check sampled configurations for collision.
 If collision-free, add as a milestone.
 Only check connections when identifying a path from initial to goal configurations.
 Test the midpoint of the longest untested segment recursively.
 Transfer milestones between trees to preserve work done if a collision is found.
4. Key Observations:
o Most local paths in a roadmap are not part of the final path.
o Testing a collision-free connection is costly.
o Shorter connections between milestones have a higher prior probability of being collision-free.
o Testing early (before necessary) is inefficient.
5. Experimental Results:
o SBL significantly reduces collision-checking time compared to full collision-checker planners.
o Convergence rates improve exponentially with higher sampling density.
o SBL achieves better performance overall.

Sampling based Roadmap

sampling-based roadmaps in motion planning:

1. Probabilistic Roadmap Method (PRM):
o PRM is a popular sampling-based path planning method.
o It constructs a roadmap by randomly sampling configurations from the configuration space.
o Nodes in the roadmap represent valid configurations, and edges connect collision-free paths
between sampled configurations1.
o PRM is effective for multiple-query motion planning, where a roadmap is built during
preprocessing to quickly respond to online queries2.
2. Smooth Transition to Single-Query Planners:
o Our novel planner combines PRM with sampling-based tree methods (e.g., Expansive Space
Trees, Rapidly-Exploring Random Trees) to achieve a smooth spectrum between multiple-query
and single-query planning.
o By exploiting this hybrid approach, our planner can solve problems that existing planners
struggle with.
o Notably, our planner is significantly more decoupled than PRM and other tree-based planners,
allowing for efficient parallelization3.
3. Parallel Implementation:
o Our planner can be used sequentially or in a powerful parallel implementation.
o It distributes well and handles high-dimensional problems that exhaust resources available to
single machines.
 Analysis Probablistic Road map

probabilistic foundations of Probabilistic Roadmap (PRM) planning:

1. Why Probabilistic?:
o PRM planners construct a simplified representation of the free space in a robot’s configuration
space.
o Instead of computing an exact representation of the collision-free subset, PRM builds a
probabilistic roadmap.
o A roadmap consists of nodes (sampled configurations) connected by collision-free edges.
o The inherent randomness in PRM arises from the probabilistic sampling measure used during
roadmap construction.
o PRM’s success depends on favorable “visibility” properties of the configuration space.
o These properties ensure efficient exploration and connectivity within the roadmap.
2. Visibility Properties:
o Good PRM performance relies on the existence of collision-free paths between sampled
configurations.
o Previous work has identified and formalized visibility properties that guarantee success with
uniform sampling measures.
o Promising directions for improving PRM planners involve inferring partial knowledge about
visibility properties from workspace geometry and roadmap construction data.
o This knowledge can guide adaptive sampling measures.
3. Sampling Source Impact:
o The choice of sampling source (pseudo-random or deterministic) has minimal impact on PRM
performance.
o The critical factor is the sampling measure itself.
o Empirical and theoretical results support these conclusions.

Control based Planning

Control-based planning in robotics involves designing algorithms and strategies to guide

robots from a starting point to a destination while avoiding obstacles. Let’s explore some key
approaches:
1. Motion Planning Algorithms:
o Graph Search Algorithms: These explore a graph representation of the environment to find a
nominal trajectory. Examples include Rapidly Exploring Random Trees
(RRT) and Probabilistic Road Maps (PRM).
o Sampling-Based Algorithms: These randomly sample the configuration space to build a
roadmap for efficient path planning.
o Interpolating Curve Algorithms: These create smooth paths by interpolating between
waypoints.
o Reaction-Based Algorithms: These react to sensory input during execution.
2. Optimal Value Reinforcement Learning:
o Algorithms like Q-learning, Deep Q-learning Network (DQN), and Dueling DQN optimize
value functions to find optimal policies.
o Double DQN and Dueling DQN improve stability and convergence.
3. Policy Gradient Algorithms:
o These directly optimize policy parameters. Examples include Actor-Critic, Advantage Actor-
Critic, and Trust Region Policy Optimization (TRPO).
4. Future Directions:
o Researchers continue to explore novel algorithms and evaluate their performance.
o The convergence speed and stability of optimal value and policy gradient methods are areas of
active research.

Multiple robots

Multiple robots and how they collaborate to achieve various tasks.

Swarm Robotics:

o Definition: Swarm robotics is an approach where large numbers of mostly simple physical
robots coordinate as a system.
o Collective Behavior: In a robot swarm, the collective behavior emerges from local interactions
between robots and their environment.
o Inspiration: It draws inspiration from social insects (like ants) and artificial swarm intelligence.
o Advantages:
 Effectiveness: A fleet of homogeneous robots outperforms a single robot in most scenarios.
 Efficiency: Heterogeneous fleets manage resources better (e.g., aerial and ground robots in
search tasks).
 Flexibility: Multi-robot systems adapt to changes (scenario, tasks, or fleet) more effectively.
 Fault Tolerance: When one robot fails, others can take over its functions.
o Challenges: Robot autonomy and human factors remain critical challenges.
 Autonomy: Robots need more capabilities to operate longer in adverse conditions.
 Human Factors: Operators face workload, situational awareness, and stress challenges1.

Multi-Robot Systems (MRS):

o MRSs are an alternative to single robots, offering advantages like effectiveness, efficiency,
flexibility, and fault tolerance.
o Challenges include autonomy, human workload, and system complexity.
o Trends focus on swarms and their applications.
o Applications span search, exploration, environmental monitoring, and more

Manipulation Planning

Manipulation planning in robotics involves designing algorithms and strategies for robots to
effectively manipulate objects. Let’s explore some key aspects:
1. Contact-Rich Manipulation Planning:
o MIT researchers have simplified manipulation planning using an AI technique
called smoothing. This technique summarizes multiple contact events into fewer
decisions, enabling efficient manipulation plans for robots1.
o The goal is to choose stable grasps, adapt grasp changes based on external forces,
and minimize regrasps in the long term2.
2. Basic Concepts:
o Pick and Place: A fundamental task where a robot picks an object from one
location and places it in another.
o Spatial Transforms: Representations for describing object poses and
transformations.
o Kinematics: Forward and differential kinematics for understanding robot motion.
o Geometric Pose Estimation: Techniques for estimating object poses using
cameras and depth sensors.
3. Hardware and Sensors:
o Robots can have different arms (position-controlled or torque-controlled) and end
effectors (hands, grippers, etc.).
o Sensors play a crucial role in perception during manipulation tasks.
4. Challenges:
o Complexity: Manipulation planning involves dealing with uncertainty, contact
dynamics, and collision avoidance.
o Real-World Constraints: Objects may be cluttered, and external forces can
change during manipulation.
o

Assembly planning

Assembly planning in robotics involves designing strategies for robots to efficiently and
accurately assemble complex structures or products. Here are some notable approaches:
1. Robotic Furniture Assembly:
o Researchers have developed frameworks for assembling furniture using robots.
o A recent study proposed a novel approach for chair assembly based on assembly instructions.
o Key components of this framework include:
 Task Templates: These templates define motion planning and control strategies for specific
assembly tasks.
 Task Compiler: Converts assembly instructions into a series of robotic tasks using predefined
templates.
 Motion Planning Optimization: Computed goal configurations avoid joint limits and
singularities.
 Control Strategies: Introduce macro- and micro-motions to handle various assembly types and
uncertainties.
o The framework was validated by successfully assembling a chair, including benchmark tasks like
pin insertion1.
2. Cooperative Task and Motion Planning:
o When multiple robots collaborate on assembly tasks, planning becomes more complex.
o Challenges include:
 Assigning tasks to robots.
 Scheduling tasks subject to temporal constraints.
 Solving collision-free motions for each manipulation.
o Researchers explore solutions for multi-arm assembly systems2.
3. Algorithmic Planning for Robotic Assembly:
o Investigations cover three broad classes of assembly problems:
 Spatial Extrusion: Planning for continuous material deposition (e.g., 3D printing).
 Pick-and-Place Assembly: Efficiently handling discrete parts.
 Robotic Assembly with Multiple Tool Changes: Addressing tool changes during assembly.