Skip to main content

Advanced AI Prompts for Chapter 3

Use these prompts with AI assistants to get help with advanced costmap configuration, planner tuning, reinforcement learning, and sim-to-real transfer.

Costmap Configuration​

Layer Configuration​

Explain the interaction between static, obstacle, and inflation layers in Nav2 costmaps.
How do they combine to produce the final cost values?

Inflation Tuning​

I have a robot with 0.5m diameter. What inflation_radius should I use?
Explain the relationship between robot size, inflation radius, and safety margins.

Custom Costmap Layers​

How do I create a custom costmap layer plugin for Nav2?
Show me the basic structure and required methods.

Global vs Local Costmap Differences​

What parameters should differ between global and local costmaps?
Explain the trade-offs in update frequency, size, and resolution.

Costmap Performance Optimization​

My costmaps are updating too slowly. What parameters affect performance?
How do I balance accuracy with computational cost?

Multi-Layer Fusion​

How does Nav2 combine multiple costmap layers?
Explain the cost combination algorithm and priority handling.

Dynamic Reconfiguration​

Can I change costmap parameters at runtime? Show me how to
dynamically reconfigure inflation radius or update frequency.

Planner Configuration​

Global Planner Comparison​

Compare A*, Dijkstra, Theta*, and Smac Planner for Nav2.
When should I use each? What are the trade-offs?

A* Heuristic Tuning​

Explain the heuristic function in A* planning.
How does it affect optimality and computation time?

DWA Parameter Tuning​

My DWA local planner is too conservative/aggressive. What parameters
control this behavior? Explain velocity limits, trajectory scoring, and obstacle costs.

TEB Planner Configuration​

When should I use TEB (Timed Elastic Band) instead of DWA?
What are the key parameters to tune?

MPPI Controller​

Explain the MPPI (Model Predictive Path Integral) controller.
What advantages does it offer? When should I use it?

Planner Switching​

Can I switch between planners at runtime based on the situation?
How would I implement adaptive planner selection?

Kinematic Constraints​

How do I configure planners to respect my robot's kinematic constraints
(max velocity, acceleration, turning radius)?

Behavior Trees​

Behavior Tree Structure​

Explain the structure of Nav2's behavior tree.
What are the main nodes and how do they coordinate navigation?

Custom Behavior Nodes​

How do I create a custom behavior tree node for Nav2?
Show me the plugin structure and registration process.

Recovery Behavior Configuration​

What recovery behaviors are available in Nav2?
How do I configure their order and parameters?

Behavior Tree Debugging​

My behavior tree isn't executing as expected. How do I debug it?
What tools or logging can help visualize execution?

State Machine vs Behavior Tree​

When should I use a behavior tree vs a traditional state machine
for robot control? What are the trade-offs?

Parallel Execution​

How do behavior trees handle parallel execution of multiple behaviors?
Show me an example of concurrent navigation and monitoring.

Reinforcement Learning​

MDP Formulation​

Help me formulate robot navigation as an MDP (Markov Decision Process).
What are the states, actions, rewards, and transition dynamics?

Reward Function Design​

Design a reward function for training a robot to navigate to a goal
while avoiding obstacles. What should be rewarded/penalized?

PPO Algorithm Explanation​

Explain the PPO (Proximal Policy Optimization) algorithm in detail.
What is the clipped objective? Why does it improve stability?

PPO vs SAC Comparison​

Compare PPO and SAC (Soft Actor-Critic) for robot control tasks.
When should I use each? What are the sample efficiency differences?

Policy Network Architecture​

What neural network architecture should I use for a navigation policy?
Explain input features, hidden layers, and output actions.

Value Function Approximation​

Explain value function approximation in RL.
What's the difference between state value (V) and action value (Q)?

Exploration Strategies​

What exploration strategies work well for robot learning?
Explain epsilon-greedy, entropy regularization, and curiosity-driven exploration.

Training Stability​

My RL training is unstable with high variance. What techniques
improve stability? Explain baseline subtraction, advantage estimation, and normalization.

Curriculum Learning​

How do I implement curriculum learning for robot tasks?
Start with simple scenarios and gradually increase difficulty.

Off-Policy vs On-Policy​

Explain the difference between on-policy (PPO) and off-policy (SAC) RL.
What are the implications for sample efficiency and stability?

Sim-to-Real Transfer​

Reality Gap Analysis​

What are the main sources of the reality gap between simulation and real robots?
How do physics, sensors, and actuators differ?

Domain Randomization Strategy​

Design a domain randomization strategy for training a navigation policy.
What parameters should I randomize? What ranges are appropriate?

System Identification​

How do I perform system identification to make my simulation more accurate?
What robot parameters should I measure and tune?

Sensor Noise Modeling​

How do I add realistic sensor noise to my Gazebo simulation?
What noise models are appropriate for cameras, LIDAR, and IMU?

Physics Simulation Accuracy​

What Gazebo physics parameters affect sim-to-real transfer?
How do I tune friction, damping, and contact parameters?

Robust Policy Training​

How do I train policies that are robust to sim-to-real differences?
Explain techniques beyond domain randomization.

ONNX Model Export​

Show me how to export a trained PyTorch/TensorFlow policy to ONNX format.
Include quantization and optimization steps.

Loading ONNX in ROS 2​

Write a ROS 2 node that loads an ONNX model and executes it for inference.
Include input preprocessing and output postprocessing.

Validation Protocol​

Design a validation protocol for testing sim-to-real transfer.
What metrics should I measure? How do I ensure safety?

Deployment Pipeline​

Outline the complete pipeline from training in simulation to deploying
on a real robot. What are the critical steps and checkpoints?

Advanced Debugging​

Costmap Visualization Analysis​

My costmap shows unexpected behavior. Walk me through systematic analysis:
What should I check in each layer? How do I isolate issues?

Planner Failure Diagnosis​

Navigation fails in specific scenarios. How do I diagnose whether
it's a global planner, local planner, or costmap issue?

TF Performance Issues​

My TF lookups are slow and causing delays. How do I optimize TF performance?
What are common bottlenecks?

Memory Leaks in Perception​

My perception node's memory usage grows over time. How do I identify
and fix memory leaks in image processing pipelines?

Real-Time Constraint Violations​

My system occasionally misses real-time deadlines. How do I profile
and optimize for consistent real-time performance?

Production Deployment​

Safety Systems​

What safety systems should I implement before deploying autonomous navigation
on a real robot? Include emergency stops, watchdogs, and fail-safes.

Monitoring & Diagnostics​

Design a monitoring system for production robot deployment.
What metrics should I track? How do I detect anomalies?

Graceful Degradation​

How do I implement graceful degradation when sensors fail or
navigation becomes unreliable? Design a fault-tolerant architecture.

Performance Benchmarking​

What benchmarks should I run to validate navigation performance?
Include metrics for accuracy, speed, safety, and robustness.

Configuration Management​

How should I manage configurations for different environments
(lab, warehouse, outdoor)? Best practices for parameter organization.

Advanced Algorithms​

Particle Filter Localization​

Explain particle filter localization in detail.
How does it differ from Kalman filter approaches?

Graph-Based SLAM​

What is graph-based SLAM? How does it differ from filter-based SLAM?
Explain pose graphs and optimization.

Semantic SLAM​

What is semantic SLAM? How does it incorporate object recognition
into mapping and localization?

Multi-Robot SLAM​

How do I extend SLAM to multiple robots? What are the challenges
in distributed mapping and localization?

Deep Learning for Perception​

How can I integrate deep learning models (object detection, segmentation)
into my perception pipeline? Architecture and performance considerations.

Research & Advanced Topics​

Imitation Learning​

Explain imitation learning for robotics. How does it differ from RL?
When should I use behavioral cloning vs inverse RL?

Meta-Learning for Adaptation​

What is meta-learning and how can it help robots adapt quickly
to new environments? Explain MAML and related approaches.

Uncertainty Quantification​

How do I quantify uncertainty in perception and planning?
Explain Bayesian approaches and ensemble methods.

Sim-to-Real via Domain Adaptation​

Explain domain adaptation techniques for sim-to-real transfer.
How do adversarial training and feature alignment help?

End-to-End Learning​

What are the trade-offs of end-to-end learning (pixels to actions)
vs modular pipelines (perception → planning → control)?

Optimization Techniques​

Hyperparameter Tuning​

What's the best approach for tuning Nav2 hyperparameters?
Should I use grid search, random search, or Bayesian optimization?

Multi-Objective Optimization​

I need to optimize for both speed and safety in navigation.
How do I handle multi-objective optimization? Pareto frontiers?

Online Learning​

Can my robot continue learning and improving during deployment?
What are safe online learning strategies?

Transfer Learning​

I trained a policy in one environment. How do I transfer it to
a different environment? What techniques preserve learned knowledge?

Integration Patterns​

Perception-Planning Integration​

Design the interface between perception and planning systems.
What information should perception provide? How should it be formatted?

Multi-Modal Sensor Fusion​

How do I fuse camera, LIDAR, and IMU data for robust perception?
Explain early fusion vs late fusion approaches.

Hierarchical Planning​

Design a hierarchical planning system with high-level task planning
and low-level motion planning. How do they communicate?

Learning-Based Planning​

How do I integrate learned policies with traditional planners?
When should the robot use learning vs classical planning?

Tips for Using These Prompts​

  • Provide Context: Mention your robot platform, sensors, and environment
  • Share Data: Include parameter files, error logs, or performance metrics
  • Ask for Trade-offs: Request analysis of different approaches
  • Request Papers: Ask for relevant research papers for deep dives
  • Iterate Solutions: Start with basic approach, then optimize

Advanced Learning Resources​

When AI assistants reference these, ask for explanations:

  • Papers: "Explain the key ideas from [paper name] in simple terms"
  • Algorithms: "Walk me through [algorithm] step by step with an example"
  • Math: "Explain the intuition behind [equation] without heavy math"
  • Code: "Show me a minimal implementation of [concept]"

Remember: Advanced topics require deep understanding. Don't hesitate to ask for multiple explanations from different angles until concepts click.

Previous: Intermediate AI Prompts