Quiz#

This quiz covers the key concepts from Lecture 13: Mapping and Navigation with Nav2. Topics include map representations, occupancy grids, the map frame and REP 105, SLAM with slam_toolbox (scan matching, pose graph, loop closure), localization with AMCL (particle filters, predict/update/resample), costmaps (global vs. local, layers, inflation, footprint), planners and controllers, the behavior tree, and the NavigateToPose action API.

Note

Instructions:

  • Answer all questions to the best of your ability.

  • Multiple choice questions have exactly one correct answer.

  • True/False questions require you to determine if the statement is correct.

  • Essay questions require short written responses (2-4 sentences).

  • Click the dropdown after each question to reveal the answer.

Multiple Choice#

Question 1

In a nav_msgs/OccupancyGrid, what does a cell value of -1 mean?

  1. The cell is free space.

  2. The cell is occupied.

  3. The cell has not been observed yet (unknown).

  4. The cell is invalid and should be ignored.

Answer

C – The cell is unknown.

OccupancyGrid uses 0 for free, 100 for occupied, and -1 for unknown. RViz2 renders unknown cells in grey.

Question 2

Which transform in the REP 105 chain is allowed to jump?

  1. base_link -> sensor frames

  2. odom -> base_link

  3. map -> odom

  4. world -> map

Answer

C – map -> odom.

The localization stack (SLAM or AMCL) corrects accumulated drift by adjusting this transform. The odom -> base_link transform must remain smooth (no jumps); the map -> odom transform absorbs the corrections.

Question 3

In slam_toolbox, what does loop closure add to the pose graph?

  1. A new node at the robot’s current pose.

  2. An edge connecting two distant nodes that observed the same area.

  3. A new occupancy grid layer.

  4. A copy of every previous scan as a fresh node.

Answer

B – An edge connecting two distant nodes.

When the robot revisits an area, scan matching between the current scan and the stored scan creates a constraint between the two corresponding nodes. The graph optimizer then realigns all poses to satisfy this new constraint, correcting accumulated drift.

Question 4

Why is AMCL called adaptive?

  1. It can run on any robot regardless of sensor configuration.

  2. It varies the number of particles based on how well-localized the robot is.

  3. It automatically chooses between a 2D and 3D map.

  4. It adapts the map resolution at runtime.

Answer

B – It varies the particle count.

When the robot is lost, AMCL uses many particles to cover the uncertainty. When the cloud has converged, it shrinks the particle set to save CPU. Standard MCL uses a fixed count.

Question 5

Which two costmaps does Nav2 maintain, and what is each used for?

  1. Static for the map; dynamic for the robot.

  2. Global (whole map, used by the planner) and local (rolling window, used by the controller).

  3. Inflation for safety; obstacle for path planning.

  4. Sensor and plan, both updated at 20 Hz.

Answer

B – Global and local.

The global costmap covers the entire map and feeds the global planner. The local costmap is a rolling window around the robot updated at sensor rate; it feeds the local controller and handles dynamic obstacles.

Question 6

In Nav2, which planner is best suited to a non-holonomic robot that must navigate tight spaces with kinematic constraints enforced from the start?

  1. NavFn (Dijkstra/A*)

  2. Smac Hybrid A*

  3. DWB

  4. Regulated Pure Pursuit

Answer

B – Smac Hybrid A*.

Smac Hybrid A* searches an SE(2) lattice that respects the robot’s turning radius, so the resulting path is physically drivable. NavFn ignores kinematics, and DWB / RPP are controllers, not planners.

Question 7

What is the role of the inflation layer in a Nav2 costmap?

  1. It increases the resolution of the costmap near obstacles.

  2. It expands lethal cells outward, creating a graded cost gradient that steers the robot away from walls.

  3. It removes obstacles that are far from the robot.

  4. It records which cells were inflated by past plans.

Answer

B – Expands lethal cells outward.

The inflation layer turns a hard binary obstacle into a smooth cost gradient. The planner then prefers paths that stay away from walls without explicitly modelling the robot’s footprint at every cell.

Question 8

In the NavigateToPose action, what does feedback provide?

  1. The final outcome of the navigation task.

  2. A periodic stream containing the current pose, distance remaining, and recovery count while the goal is active.

  3. A copy of the goal pose echoed back to the client.

  4. The full planned path on every tick.

Answer

B – A periodic stream of progress information.

Feedback includes current_pose, navigation_time, estimated_time_remaining, distance_remaining, and number_of_recoveries. The result is sent only once when the goal terminates.

True/False#

Question 9

Setting set_initial_pose: true in AMCL’s parameters and providing initial_pose_x, initial_pose_y, and initial_pose_a is a valid alternative to clicking 2D Pose Estimate in RViz2.

Answer

True – Both methods deliver the same initial pose to AMCL. The parameter approach is reproducible and well suited to scripted tests; the RViz2 button is convenient interactively.

Question 10

The robot’s circumscribed radius is the radius of the largest circle that fits inside the footprint.

Answer

False – That definition describes the inscribed radius. The circumscribed radius is the smallest circle that encloses the footprint.

Question 11

slam_toolbox can be used to localize against a previously saved map without modifying it.

Answer

True – slam_toolbox has a localization mode that loads an existing serialized graph and runs scan matching against it without adding new nodes, similar in role to AMCL.

Question 12

In Nav2, the global planner is called every control cycle to produce a fresh path.

Answer

False – The planner is called once per goal request and replanned only when needed (e.g., the path becomes blocked or a new goal arrives). The local controller runs every control cycle, typically at 20 Hz.

Question 13

When using BasicNavigator.followWaypoints(poses), you must manually create your own NavigateToPose action client.

Answer

False – BasicNavigator wraps the underlying action clients. You only construct the list of PoseStamped waypoints and call followWaypoints; the action plumbing is hidden.

Essay Questions#

Question 14

Explain how scan matching, the pose graph, and loop closure work together in slam_toolbox to produce a globally consistent map. Why is loop closure essential after long traversals?

Answer

Scan matching aligns consecutive LiDAR scans to estimate the relative motion between two robot poses; each result becomes an edge in the pose graph. The pose graph stores these edges as constraints between nodes that represent successive robot poses along the trajectory. Sequential edges accumulate small errors over time, so without correction the map would shear and overlap. Loop closure detects when the robot revisits a previously seen area, runs scan matching between the current and stored scans, and adds a new edge connecting two distant nodes. The graph optimizer then adjusts every node simultaneously to minimize the total constraint error, redistributing accumulated drift across the entire trajectory and producing a globally consistent occupancy grid.

Question 15

Describe the predict / update / resample cycle of AMCL’s particle filter. What role does each step play in keeping the pose estimate accurate?

Answer

In predict, every particle is shifted by the robot’s measured motion plus a small amount of noise drawn from the motion model. This propagates the estimate forward in time and spreads the particles slightly to reflect uncertainty in the motion. In update, each particle is scored by casting a virtual LiDAR from its pose into the map and comparing the predicted ranges to the actual sensor reading; particles that explain the data well gain weight. In resample, a new set of particles is drawn with replacement, with probability proportional to weight, so high-weight hypotheses are duplicated and low-weight ones are dropped. Repeated cycles concentrate the cloud around poses that consistently explain motion and observation, giving a converged pose estimate.

Question 16

Compare the global costmap and the local costmap in Nav2. Why does Nav2 maintain two separate costmaps instead of one?

Answer

The global costmap covers the entire map and is used by the global planner; it is built primarily from the static map (and slow-changing obstacles), updated at a low rate, and supports long-range route planning. The local costmap is a rolling window centred on the robot, updated at sensor rate from live LiDAR / depth data, and is used by the local controller to react to dynamic obstacles such as a person stepping into a corridor. Splitting them means each costmap can be sized and tuned for its own role: the global one is large but cheap to update; the local one is small but high-frequency and aggressive about new obstacles. Trying to use a single costmap would force a bad tradeoff between coverage, latency, and CPU cost.

Question 17

In the NavigateToPose action interface, distinguish goal, feedback, and result, and explain why this three-part structure is appropriate for navigation.

Answer

The goal is the target PoseStamped in the map frame, sent once when the client requests navigation. Feedback is published periodically while the task is running and contains live progress information – current pose, distance remaining, time elapsed, and recovery count. The result is sent exactly once when the task terminates and conveys the final status code (SUCCEEDED, CANCELED, or FAILED). This three-part structure fits navigation because the task is long-running (so a one-shot service would either block or time out), the client typically wants progress updates without polling, and the client must be able to cancel the goal mid-execution if a higher-priority task arrives.