Planning-based prediction for pedestrians
read more
Citations
Social LSTM: Human Trajectory Prediction in Crowded Spaces
Activity forecasting
CHOMP: Covariant Hamiltonian optimization for motion planning
Human-aware robot navigation: A survey
Learning Social Etiquette: Human Trajectory Understanding In Crowded Scenes
References
Reinforcement Learning: An Introduction
A note on two problems in connexion with graphs
Introduction to Reinforcement Learning
New Results in Linear Filtering and Prediction Theory
Apprenticeship learning via inverse reinforcement learning
Related Papers (5)
Frequently Asked Questions (15)
Q2. What future works have the authors mentioned in the paper "Planning-based prediction for pedestrians" ?
The authors have presented a novel approach for predicting future pedestrian trajectories using a soft-max version of goal-based planning. The authors additionally showed the usefulness of this approach for planning hindrance-sensitive routes using a novel incremental path planner. In future work, the authors plan to explicitly model interactions between people so that they can better predict movements in crowded environments.
Q3. What are the main features that influence movement in many environments?
In many environments, the relevant features that influence movement change frequently – furniture is moved in indoor environments, the locations of parked vehicles are dynamic in urban environments, and weather conditions influence natural environments with muddy, icy, or dry conditions.
Q4. What is the entropy of the distribution of trajectories?
By maximizing the entropy of the distribution of trajectories, H(Pζ) = − ∑ ζ P (ζ) logP (ζ) subject to the constraint of matching the reward of the person’s behavior in expectation [1], the authors obtain a distribution over trajectories [18].
Q5. What is the way to predict a set of observed trajectories?
For prescriptive MDP applications, the reward values for actions in the MDP are often engineered to produce appropriate behavior, however for their prediction purposes,we would like to find the reward values that best predict a set of observed trajectories, {ζ̃i}.
Q6. What is the method for determining the optimal quantities?
(2)The value iteration algorithm produces these optimal values by alternately applying Equations 1 and 2 as update rules until the values converge.
Q7. How do the authors smooth the probability of a given location to nearby cells?
The authors smooth this probability to nearby cells using the Manhattan distance (dist(a, b)) and also add probability (P0) for previously unvisited locations to avoid overfitting, yielding: P (dest x) ∝ P0 + ∑ goals g e−dist(x,g).
Q8. How can the authors improve the computational gain of this technique?
The authors can further improve the computational gain of this technique by using efficient replanners such as D* and its variants [4] in the inner loop.
Q9. How do the authors achieve intelligent adaptive robot behavior?
By re-running this iterative replanner every 0.25 seconds using updated predictions of pedestrian motion, the authors can achieve intelligent adaptive robot behavior that anticipates where a pedestrian is heading and maneuvers well in advance to implement efficient avoidance.
Q10. What is the way to determine the optimal trajectories?
Trajectories with a very high reward (low cost) are exponentially more preferable to low reward (high cost) trajectories, and trajectories with equal reward are equally probable.
Q11. What is the procedure for integrating predictive pedestrian models?
Algorithm 1 Incorporating predictive pedestrian models via predictive planning1: procedure PREDICTIVEPLANNING(σ > 0, α > 0, {Ds,t}, Dthresh) 2: Initialize cost map to prior navigational costs c0(s).
Q12. How do the authors plan to model interactions between people?
In future work, the authors plan to explicitly model interactions between people so that the authors can better predict movements in crowded environments.
Q13. What is the cost-to-go value of the iteratively constructed cost map?
In practice, the authors use the final cost-to-go values of the iteratively constructed cost map to implement a policy that chooses a good action from a predefined collection of actions.
Q14. How can the authors predict human behavior in crowded environments?
As the authors have shown, the feature-based cost function learned using this approach allows accurate generalization to changes in the environment.
Q15. How do the authors replace the maximum of the Bellman equations with a softmax?
The authors accomplish this by replacing the maximum of the Bellman equations with a soft-maximum function, softmaxx f(x) = log ∑ x e f(x).